mailinfo: support format=flowed
authorRené Scharfe <l.s.r@web.de>
Sat, 25 Aug 2018 21:50:32 +0000 (23:50 +0200)
committerJunio C Hamano <gitster@pobox.com>
Wed, 29 Aug 2018 20:05:35 +0000 (13:05 -0700)
Add best-effort support for patches sent using format=flowed (RFC 3676).
Remove leading spaces ("unstuff"), remove soft line breaks (indicated
by space + newline), but leave the signature separator (dash dash space
newline) alone.

Warn in git am when encountering a format=flowed patch, because any
trailing spaces would most probably be lost, as the sending MUA is
encouraged to remove them when preparing the email.

Provide a test patch formatted by Mozilla Thunderbird 60 using its
default configuration.  It reuses the contents of the file mailinfo.c
before and after this patch.

Signed-off-by: Rene Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
builtin/am.c
mailinfo.c
mailinfo.h
t/t4256-am-format-flowed.sh [new file with mode: 0755]
t/t4256/1/mailinfo.c [new file with mode: 0644]
t/t4256/1/mailinfo.c.orig [new file with mode: 0644]
t/t4256/1/patch [new file with mode: 0644]

index 2fc2d1e..30c87c1 100644 (file)
@@ -1243,6 +1243,10 @@ static int parse_mail(struct am_state *state, const char *mail)
        fclose(mi.input);
        fclose(mi.output);
 
+       if (mi.format_flowed)
+               warning(_("Patch sent with format=flowed; "
+                         "space at the end of lines might be lost."));
+
        /* Extract message and author information */
        fp = xfopen(am_path(state, "info"), "r");
        while (!strbuf_getline_lf(&sb, fp)) {
index 3281a37..b395adb 100644 (file)
@@ -237,11 +237,22 @@ static int slurp_attr(const char *line, const char *name, struct strbuf *attr)
        return 1;
 }
 
+static int has_attr_value(const char *line, const char *name, const char *value)
+{
+       struct strbuf sb = STRBUF_INIT;
+       int rc = slurp_attr(line, name, &sb) && !strcasecmp(sb.buf, value);
+       strbuf_release(&sb);
+       return rc;
+}
+
 static void handle_content_type(struct mailinfo *mi, struct strbuf *line)
 {
        struct strbuf *boundary = xmalloc(sizeof(struct strbuf));
        strbuf_init(boundary, line->len);
 
+       mi->format_flowed = has_attr_value(line->buf, "format=", "flowed");
+       mi->delsp = has_attr_value(line->buf, "delsp=", "yes");
+
        if (slurp_attr(line->buf, "boundary=", boundary)) {
                strbuf_insert(boundary, 0, "--", 2);
                if (++mi->content_top >= &mi->content[MAX_BOUNDARIES]) {
@@ -964,6 +975,52 @@ again:
        return 1;
 }
 
+static void handle_filter_flowed(struct mailinfo *mi, struct strbuf *line,
+                                struct strbuf *prev)
+{
+       size_t len = line->len;
+       const char *rest;
+
+       if (!mi->format_flowed) {
+               handle_filter(mi, line);
+               return;
+       }
+
+       if (line->buf[len - 1] == '\n') {
+               len--;
+               if (len && line->buf[len - 1] == '\r')
+                       len--;
+       }
+
+       /* Keep signature separator as-is. */
+       if (skip_prefix(line->buf, "-- ", &rest) && rest - line->buf == len) {
+               if (prev->len) {
+                       handle_filter(mi, prev);
+                       strbuf_reset(prev);
+               }
+               handle_filter(mi, line);
+               return;
+       }
+
+       /* Unstuff space-stuffed line. */
+       if (len && line->buf[0] == ' ') {
+               strbuf_remove(line, 0, 1);
+               len--;
+       }
+
+       /* Save flowed line for later, but without the soft line break. */
+       if (len && line->buf[len - 1] == ' ') {
+               strbuf_add(prev, line->buf, len - !!mi->delsp);
+               return;
+       }
+
+       /* Prepend any previous partial lines */
+       strbuf_insert(line, 0, prev->buf, prev->len);
+       strbuf_reset(prev);
+
+       handle_filter(mi, line);
+}
+
 static void handle_body(struct mailinfo *mi, struct strbuf *line)
 {
        struct strbuf prev = STRBUF_INIT;
@@ -1012,7 +1069,7 @@ static void handle_body(struct mailinfo *mi, struct strbuf *line)
                                                strbuf_addbuf(&prev, sb);
                                                break;
                                        }
-                               handle_filter(mi, sb);
+                               handle_filter_flowed(mi, sb, &prev);
                        }
                        /*
                         * The partial chunk is saved in "prev" and will be
@@ -1022,13 +1079,16 @@ static void handle_body(struct mailinfo *mi, struct strbuf *line)
                        break;
                }
                default:
-                       handle_filter(mi, line);
+                       handle_filter_flowed(mi, line, &prev);
                }
 
                if (mi->input_error)
                        break;
        } while (!strbuf_getwholeline(line, mi->input, '\n'));
 
+       if (prev.len)
+               handle_filter(mi, &prev);
+
        flush_inbody_header_accum(mi);
 
 handle_body_out:
index 04a2535..67af28d 100644 (file)
@@ -20,6 +20,8 @@ struct mailinfo {
        struct strbuf *content[MAX_BOUNDARIES];
        struct strbuf **content_top;
        struct strbuf charset;
+       unsigned int format_flowed:1;
+       unsigned int delsp:1;
        char *message_id;
        enum  {
                TE_DONTCARE, TE_QP, TE_BASE64
diff --git a/t/t4256-am-format-flowed.sh b/t/t4256-am-format-flowed.sh
new file mode 100755 (executable)
index 0000000..6340310
--- /dev/null
@@ -0,0 +1,19 @@
+#!/bin/sh
+
+test_description='test format=flowed support of git am'
+
+. ./test-lib.sh
+
+test_expect_success 'setup' '
+       cp "$TEST_DIRECTORY/t4256/1/mailinfo.c.orig" mailinfo.c &&
+       git add mailinfo.c &&
+       git commit -m initial
+'
+
+test_expect_success 'am with format=flowed' '
+       git am <"$TEST_DIRECTORY/t4256/1/patch" >stdout 2>stderr &&
+       test_i18ngrep "warning: Patch sent with format=flowed" stderr &&
+       test_cmp "$TEST_DIRECTORY/t4256/1/mailinfo.c" mailinfo.c
+'
+
+test_done
diff --git a/t/t4256/1/mailinfo.c b/t/t4256/1/mailinfo.c
new file mode 100644 (file)
index 0000000..b395adb
--- /dev/null
@@ -0,0 +1,1245 @@
+#include "cache.h"
+#include "config.h"
+#include "utf8.h"
+#include "strbuf.h"
+#include "mailinfo.h"
+
+static void cleanup_space(struct strbuf *sb)
+{
+       size_t pos, cnt;
+       for (pos = 0; pos < sb->len; pos++) {
+               if (isspace(sb->buf[pos])) {
+                       sb->buf[pos] = ' ';
+                       for (cnt = 0; isspace(sb->buf[pos + cnt + 1]); cnt++);
+                       strbuf_remove(sb, pos + 1, cnt);
+               }
+       }
+}
+
+static void get_sane_name(struct strbuf *out, struct strbuf *name, struct strbuf *email)
+{
+       struct strbuf *src = name;
+       if (name->len < 3 || 60 < name->len || strchr(name->buf, '@') ||
+               strchr(name->buf, '<') || strchr(name->buf, '>'))
+               src = email;
+       else if (name == out)
+               return;
+       strbuf_reset(out);
+       strbuf_addbuf(out, src);
+}
+
+static void parse_bogus_from(struct mailinfo *mi, const struct strbuf *line)
+{
+       /* John Doe <johndoe> */
+
+       char *bra, *ket;
+       /* This is fallback, so do not bother if we already have an
+        * e-mail address.
+        */
+       if (mi->email.len)
+               return;
+
+       bra = strchr(line->buf, '<');
+       if (!bra)
+               return;
+       ket = strchr(bra, '>');
+       if (!ket)
+               return;
+
+       strbuf_reset(&mi->email);
+       strbuf_add(&mi->email, bra + 1, ket - bra - 1);
+
+       strbuf_reset(&mi->name);
+       strbuf_add(&mi->name, line->buf, bra - line->buf);
+       strbuf_trim(&mi->name);
+       get_sane_name(&mi->name, &mi->name, &mi->email);
+}
+
+static const char *unquote_comment(struct strbuf *outbuf, const char *in)
+{
+       int c;
+       int take_next_literally = 0;
+
+       strbuf_addch(outbuf, '(');
+
+       while ((c = *in++) != 0) {
+               if (take_next_literally == 1) {
+                       take_next_literally = 0;
+               } else {
+                       switch (c) {
+                       case '\\':
+                               take_next_literally = 1;
+                               continue;
+                       case '(':
+                               in = unquote_comment(outbuf, in);
+                               continue;
+                       case ')':
+                               strbuf_addch(outbuf, ')');
+                               return in;
+                       }
+               }
+
+               strbuf_addch(outbuf, c);
+       }
+
+       return in;
+}
+
+static const char *unquote_quoted_string(struct strbuf *outbuf, const char *in)
+{
+       int c;
+       int take_next_literally = 0;
+
+       while ((c = *in++) != 0) {
+               if (take_next_literally == 1) {
+                       take_next_literally = 0;
+               } else {
+                       switch (c) {
+                       case '\\':
+                               take_next_literally = 1;
+                               continue;
+                       case '"':
+                               return in;
+                       }
+               }
+
+               strbuf_addch(outbuf, c);
+       }
+
+       return in;
+}
+
+static void unquote_quoted_pair(struct strbuf *line)
+{
+       struct strbuf outbuf;
+       const char *in = line->buf;
+       int c;
+
+       strbuf_init(&outbuf, line->len);
+
+       while ((c = *in++) != 0) {
+               switch (c) {
+               case '"':
+                       in = unquote_quoted_string(&outbuf, in);
+                       continue;
+               case '(':
+                       in = unquote_comment(&outbuf, in);
+                       continue;
+               }
+
+               strbuf_addch(&outbuf, c);
+       }
+
+       strbuf_swap(&outbuf, line);
+       strbuf_release(&outbuf);
+
+}
+
+static void handle_from(struct mailinfo *mi, const struct strbuf *from)
+{
+       char *at;
+       size_t el;
+       struct strbuf f;
+
+       strbuf_init(&f, from->len);
+       strbuf_addbuf(&f, from);
+
+       unquote_quoted_pair(&f);
+
+       at = strchr(f.buf, '@');
+       if (!at) {
+               parse_bogus_from(mi, from);
+               goto out;
+       }
+
+       /*
+        * If we already have one email, don't take any confusing lines
+        */
+       if (mi->email.len && strchr(at + 1, '@'))
+               goto out;
+
+       /* Pick up the string around '@', possibly delimited with <>
+        * pair; that is the email part.
+        */
+       while (at > f.buf) {
+               char c = at[-1];
+               if (isspace(c))
+                       break;
+               if (c == '<') {
+                       at[-1] = ' ';
+                       break;
+               }
+               at--;
+       }
+       el = strcspn(at, " \n\t\r\v\f>");
+       strbuf_reset(&mi->email);
+       strbuf_add(&mi->email, at, el);
+       strbuf_remove(&f, at - f.buf, el + (at[el] ? 1 : 0));
+
+       /* The remainder is name.  It could be
+        *
+        * - "John Doe <john.doe@xz>"                   (a), or
+        * - "john.doe@xz (John Doe)"                   (b), or
+        * - "John (zzz) Doe <john.doe@xz> (Comment)"   (c)
+        *
+        * but we have removed the email part, so
+        *
+        * - remove extra spaces which could stay after email (case 'c'), and
+        * - trim from both ends, possibly removing the () pair at the end
+        *   (cases 'a' and 'b').
+        */
+       cleanup_space(&f);
+       strbuf_trim(&f);
+       if (f.buf[0] == '(' && f.len && f.buf[f.len - 1] == ')') {
+               strbuf_remove(&f, 0, 1);
+               strbuf_setlen(&f, f.len - 1);
+       }
+
+       get_sane_name(&mi->name, &f, &mi->email);
+out:
+       strbuf_release(&f);
+}
+
+static void handle_header(struct strbuf **out, const struct strbuf *line)
+{
+       if (!*out) {
+               *out = xmalloc(sizeof(struct strbuf));
+               strbuf_init(*out, line->len);
+       } else
+               strbuf_reset(*out);
+
+       strbuf_addbuf(*out, line);
+}
+
+/* NOTE NOTE NOTE.  We do not claim we do full MIME.  We just attempt
+ * to have enough heuristics to grok MIME encoded patches often found
+ * on our mailing lists.  For example, we do not even treat header lines
+ * case insensitively.
+ */
+
+static int slurp_attr(const char *line, const char *name, struct strbuf *attr)
+{
+       const char *ends, *ap = strcasestr(line, name);
+       size_t sz;
+
+       strbuf_setlen(attr, 0);
+       if (!ap)
+               return 0;
+       ap += strlen(name);
+       if (*ap == '"') {
+               ap++;
+               ends = "\"";
+       }
+       else
+               ends = "; \t";
+       sz = strcspn(ap, ends);
+       strbuf_add(attr, ap, sz);
+       return 1;
+}
+
+static int has_attr_value(const char *line, const char *name, const char *value)
+{
+       struct strbuf sb = STRBUF_INIT;
+       int rc = slurp_attr(line, name, &sb) && !strcasecmp(sb.buf, value);
+       strbuf_release(&sb);
+       return rc;
+}
+
+static void handle_content_type(struct mailinfo *mi, struct strbuf *line)
+{
+       struct strbuf *boundary = xmalloc(sizeof(struct strbuf));
+       strbuf_init(boundary, line->len);
+
+       mi->format_flowed = has_attr_value(line->buf, "format=", "flowed");
+       mi->delsp = has_attr_value(line->buf, "delsp=", "yes");
+
+       if (slurp_attr(line->buf, "boundary=", boundary)) {
+               strbuf_insert(boundary, 0, "--", 2);
+               if (++mi->content_top >= &mi->content[MAX_BOUNDARIES]) {
+                       error("Too many boundaries to handle");
+                       mi->input_error = -1;
+                       mi->content_top = &mi->content[MAX_BOUNDARIES] - 1;
+                       return;
+               }
+               *(mi->content_top) = boundary;
+               boundary = NULL;
+       }
+       slurp_attr(line->buf, "charset=", &mi->charset);
+
+       if (boundary) {
+               strbuf_release(boundary);
+               free(boundary);
+       }
+}
+
+static void handle_content_transfer_encoding(struct mailinfo *mi,
+                                            const struct strbuf *line)
+{
+       if (strcasestr(line->buf, "base64"))
+               mi->transfer_encoding = TE_BASE64;
+       else if (strcasestr(line->buf, "quoted-printable"))
+               mi->transfer_encoding = TE_QP;
+       else
+               mi->transfer_encoding = TE_DONTCARE;
+}
+
+static int is_multipart_boundary(struct mailinfo *mi, const struct strbuf *line)
+{
+       struct strbuf *content_top = *(mi->content_top);
+
+       return ((content_top->len <= line->len) &&
+               !memcmp(line->buf, content_top->buf, content_top->len));
+}
+
+static void cleanup_subject(struct mailinfo *mi, struct strbuf *subject)
+{
+       size_t at = 0;
+
+       while (at < subject->len) {
+               char *pos;
+               size_t remove;
+
+               switch (subject->buf[at]) {
+               case 'r': case 'R':
+                       if (subject->len <= at + 3)
+                               break;
+                       if ((subject->buf[at + 1] == 'e' ||
+                            subject->buf[at + 1] == 'E') &&
+                           subject->buf[at + 2] == ':') {
+                               strbuf_remove(subject, at, 3);
+                               continue;
+                       }
+                       at++;
+                       break;
+               case ' ': case '\t': case ':':
+                       strbuf_remove(subject, at, 1);
+                       continue;
+               case '[':
+                       pos = strchr(subject->buf + at, ']');
+                       if (!pos)
+                               break;
+                       remove = pos - subject->buf + at + 1;
+                       if (!mi->keep_non_patch_brackets_in_subject ||
+                           (7 <= remove &&
+                            memmem(subject->buf + at, remove, "PATCH", 5)))
+                               strbuf_remove(subject, at, remove);
+                       else {
+                               at += remove;
+                               /*
+                                * If the input had a space after the ], keep
+                                * it.  We don't bother with finding the end of
+                                * the space, since we later normalize it
+                                * anyway.
+                                */
+                               if (isspace(subject->buf[at]))
+                                       at += 1;
+                       }
+                       continue;
+               }
+               break;
+       }
+       strbuf_trim(subject);
+}
+
+#define MAX_HDR_PARSED 10
+static const char *header[MAX_HDR_PARSED] = {
+       "From","Subject","Date",
+};
+
+static inline int cmp_header(const struct strbuf *line, const char *hdr)
+{
+       int len = strlen(hdr);
+       return !strncasecmp(line->buf, hdr, len) && line->len > len &&
+                       line->buf[len] == ':' && isspace(line->buf[len + 1]);
+}
+
+static int is_format_patch_separator(const char *line, int len)
+{
+       static const char SAMPLE[] =
+               "From e6807f3efca28b30decfecb1732a56c7db1137ee Mon Sep 17 00:00:00 2001\n";
+       const char *cp;
+
+       if (len != strlen(SAMPLE))
+               return 0;
+       if (!skip_prefix(line, "From ", &cp))
+               return 0;
+       if (strspn(cp, "0123456789abcdef") != 40)
+               return 0;
+       cp += 40;
+       return !memcmp(SAMPLE + (cp - line), cp, strlen(SAMPLE) - (cp - line));
+}
+
+static struct strbuf *decode_q_segment(const struct strbuf *q_seg, int rfc2047)
+{
+       const char *in = q_seg->buf;
+       int c;
+       struct strbuf *out = xmalloc(sizeof(struct strbuf));
+       strbuf_init(out, q_seg->len);
+
+       while ((c = *in++) != 0) {
+               if (c == '=') {
+                       int ch, d = *in;
+                       if (d == '\n' || !d)
+                               break; /* drop trailing newline */
+                       ch = hex2chr(in);
+                       if (ch >= 0) {
+                               strbuf_addch(out, ch);
+                               in += 2;
+                               continue;
+                       }
+                       /* garbage -- fall through */
+               }
+               if (rfc2047 && c == '_') /* rfc2047 4.2 (2) */
+                       c = 0x20;
+               strbuf_addch(out, c);
+       }
+       return out;
+}
+
+static struct strbuf *decode_b_segment(const struct strbuf *b_seg)
+{
+       /* Decode in..ep, possibly in-place to ot */
+       int c, pos = 0, acc = 0;
+       const char *in = b_seg->buf;
+       struct strbuf *out = xmalloc(sizeof(struct strbuf));
+       strbuf_init(out, b_seg->len);
+
+       while ((c = *in++) != 0) {
+               if (c == '+')
+                       c = 62;
+               else if (c == '/')
+                       c = 63;
+               else if ('A' <= c && c <= 'Z')
+                       c -= 'A';
+               else if ('a' <= c && c <= 'z')
+                       c -= 'a' - 26;
+               else if ('0' <= c && c <= '9')
+                       c -= '0' - 52;
+               else
+                       continue; /* garbage */
+               switch (pos++) {
+               case 0:
+                       acc = (c << 2);
+                       break;
+               case 1:
+                       strbuf_addch(out, (acc | (c >> 4)));
+                       acc = (c & 15) << 4;
+                       break;
+               case 2:
+                       strbuf_addch(out, (acc | (c >> 2)));
+                       acc = (c & 3) << 6;
+                       break;
+               case 3:
+                       strbuf_addch(out, (acc | c));
+                       acc = pos = 0;
+                       break;
+               }
+       }
+       return out;
+}
+
+static int convert_to_utf8(struct mailinfo *mi,
+                          struct strbuf *line, const char *charset)
+{
+       char *out;
+
+       if (!mi->metainfo_charset || !charset || !*charset)
+               return 0;
+
+       if (same_encoding(mi->metainfo_charset, charset))
+               return 0;
+       out = reencode_string(line->buf, mi->metainfo_charset, charset);
+       if (!out) {
+               mi->input_error = -1;
+               return error("cannot convert from %s to %s",
+                            charset, mi->metainfo_charset);
+       }
+       strbuf_attach(line, out, strlen(out), strlen(out));
+       return 0;
+}
+
+static void decode_header(struct mailinfo *mi, struct strbuf *it)
+{
+       char *in, *ep, *cp;
+       struct strbuf outbuf = STRBUF_INIT, *dec;
+       struct strbuf charset_q = STRBUF_INIT, piecebuf = STRBUF_INIT;
+       int found_error = 1; /* pessimism */
+
+       in = it->buf;
+       while (in - it->buf <= it->len && (ep = strstr(in, "=?")) != NULL) {
+               int encoding;
+               strbuf_reset(&charset_q);
+               strbuf_reset(&piecebuf);
+
+               if (in != ep) {
+                       /*
+                        * We are about to process an encoded-word
+                        * that begins at ep, but there is something
+                        * before the encoded word.
+                        */
+                       char *scan;
+                       for (scan = in; scan < ep; scan++)
+                               if (!isspace(*scan))
+                                       break;
+
+                       if (scan != ep || in == it->buf) {
+                               /*
+                                * We should not lose that "something",
+                                * unless we have just processed an
+                                * encoded-word, and there is only LWS
+                                * before the one we are about to process.
+                                */
+                               strbuf_add(&outbuf, in, ep - in);
+                       }
+               }
+               /* E.g.
+                * ep : "=?iso-2022-jp?B?GyR...?= foo"
+                * ep : "=?ISO-8859-1?Q?Foo=FCbar?= baz"
+                */
+               ep += 2;
+
+               if (ep - it->buf >= it->len || !(cp = strchr(ep, '?')))
+                       goto release_return;
+
+               if (cp + 3 - it->buf > it->len)
+                       goto release_return;
+               strbuf_add(&charset_q, ep, cp - ep);
+
+               encoding = cp[1];
+               if (!encoding || cp[2] != '?')
+                       goto release_return;
+               ep = strstr(cp + 3, "?=");
+               if (!ep)
+                       goto release_return;
+               strbuf_add(&piecebuf, cp + 3, ep - cp - 3);
+               switch (tolower(encoding)) {
+               default:
+                       goto release_return;
+               case 'b':
+                       dec = decode_b_segment(&piecebuf);
+                       break;
+               case 'q':
+                       dec = decode_q_segment(&piecebuf, 1);
+                       break;
+               }
+               if (convert_to_utf8(mi, dec, charset_q.buf))
+                       goto release_return;
+
+               strbuf_addbuf(&outbuf, dec);
+               strbuf_release(dec);
+               free(dec);
+               in = ep + 2;
+       }
+       strbuf_addstr(&outbuf, in);
+       strbuf_reset(it);
+       strbuf_addbuf(it, &outbuf);
+       found_error = 0;
+release_return:
+       strbuf_release(&outbuf);
+       strbuf_release(&charset_q);
+       strbuf_release(&piecebuf);
+
+       if (found_error)
+               mi->input_error = -1;
+}
+
+static int check_header(struct mailinfo *mi,
+                       const struct strbuf *line,
+                       struct strbuf *hdr_data[], int overwrite)
+{
+       int i, ret = 0, len;
+       struct strbuf sb = STRBUF_INIT;
+
+       /* search for the interesting parts */
+       for (i = 0; header[i]; i++) {
+               int len = strlen(header[i]);
+               if ((!hdr_data[i] || overwrite) && cmp_header(line, header[i])) {
+                       /* Unwrap inline B and Q encoding, and optionally
+                        * normalize the meta information to utf8.
+                        */
+                       strbuf_add(&sb, line->buf + len + 2, line->len - len - 2);
+                       decode_header(mi, &sb);
+                       handle_header(&hdr_data[i], &sb);
+                       ret = 1;
+                       goto check_header_out;
+               }
+       }
+
+       /* Content stuff */
+       if (cmp_header(line, "Content-Type")) {
+               len = strlen("Content-Type: ");
+               strbuf_add(&sb, line->buf + len, line->len - len);
+               decode_header(mi, &sb);
+               strbuf_insert(&sb, 0, "Content-Type: ", len);
+               handle_content_type(mi, &sb);
+               ret = 1;
+               goto check_header_out;
+       }
+       if (cmp_header(line, "Content-Transfer-Encoding")) {
+               len = strlen("Content-Transfer-Encoding: ");
+               strbuf_add(&sb, line->buf + len, line->len - len);
+               decode_header(mi, &sb);
+               handle_content_transfer_encoding(mi, &sb);
+               ret = 1;
+               goto check_header_out;
+       }
+       if (cmp_header(line, "Message-Id")) {
+               len = strlen("Message-Id: ");
+               strbuf_add(&sb, line->buf + len, line->len - len);
+               decode_header(mi, &sb);
+               if (mi->add_message_id)
+                       mi->message_id = strbuf_detach(&sb, NULL);
+               ret = 1;
+               goto check_header_out;
+       }
+
+check_header_out:
+       strbuf_release(&sb);
+       return ret;
+}
+
+/*
+ * Returns 1 if the given line or any line beginning with the given line is an
+ * in-body header (that is, check_header will succeed when passed
+ * mi->s_hdr_data).
+ */
+static int is_inbody_header(const struct mailinfo *mi,
+                           const struct strbuf *line)
+{
+       int i;
+       for (i = 0; header[i]; i++)
+               if (!mi->s_hdr_data[i] && cmp_header(line, header[i]))
+                       return 1;
+       return 0;
+}
+
+static void decode_transfer_encoding(struct mailinfo *mi, struct strbuf *line)
+{
+       struct strbuf *ret;
+
+       switch (mi->transfer_encoding) {
+       case TE_QP:
+               ret = decode_q_segment(line, 0);
+               break;
+       case TE_BASE64:
+               ret = decode_b_segment(line);
+               break;
+       case TE_DONTCARE:
+       default:
+               return;
+       }
+       strbuf_reset(line);
+       strbuf_addbuf(line, ret);
+       strbuf_release(ret);
+       free(ret);
+}
+
+static inline int patchbreak(const struct strbuf *line)
+{
+       size_t i;
+
+       /* Beginning of a "diff -" header? */
+       if (starts_with(line->buf, "diff -"))
+               return 1;
+
+       /* CVS "Index: " line? */
+       if (starts_with(line->buf, "Index: "))
+               return 1;
+
+       /*
+        * "--- <filename>" starts patches without headers
+        * "---<sp>*" is a manual separator
+        */
+       if (line->len < 4)
+               return 0;
+
+       if (starts_with(line->buf, "---")) {
+               /* space followed by a filename? */
+               if (line->buf[3] == ' ' && !isspace(line->buf[4]))
+                       return 1;
+               /* Just whitespace? */
+               for (i = 3; i < line->len; i++) {
+                       unsigned char c = line->buf[i];
+                       if (c == '\n')
+                               return 1;
+                       if (!isspace(c))
+                               break;
+               }
+               return 0;
+       }
+       return 0;
+}
+
+static int is_scissors_line(const char *line)
+{
+       const char *c;
+       int scissors = 0, gap = 0;
+       const char *first_nonblank = NULL, *last_nonblank = NULL;
+       int visible, perforation = 0, in_perforation = 0;
+
+       for (c = line; *c; c++) {
+               if (isspace(*c)) {
+                       if (in_perforation) {
+                               perforation++;
+                               gap++;
+                       }
+                       continue;
+               }
+               last_nonblank = c;
+               if (first_nonblank == NULL)
+                       first_nonblank = c;
+               if (*c == '-') {
+                       in_perforation = 1;
+                       perforation++;
+                       continue;
+               }
+               if ((!memcmp(c, ">8", 2) || !memcmp(c, "8<", 2) ||
+                    !memcmp(c, ">%", 2) || !memcmp(c, "%<", 2))) {
+                       in_perforation = 1;
+                       perforation += 2;
+                       scissors += 2;
+                       c++;
+                       continue;
+               }
+               in_perforation = 0;
+       }
+
+       /*
+        * The mark must be at least 8 bytes long (e.g. "-- >8 --").
+        * Even though there can be arbitrary cruft on the same line
+        * (e.g. "cut here"), in order to avoid misidentification, the
+        * perforation must occupy more than a third of the visible
+        * width of the line, and dashes and scissors must occupy more
+        * than half of the perforation.
+        */
+
+       if (first_nonblank && last_nonblank)
+               visible = last_nonblank - first_nonblank + 1;
+       else
+               visible = 0;
+       return (scissors && 8 <= visible &&
+               visible < perforation * 3 &&
+               gap * 2 < perforation);
+}
+
+static void flush_inbody_header_accum(struct mailinfo *mi)
+{
+       if (!mi->inbody_header_accum.len)
+               return;
+       if (!check_header(mi, &mi->inbody_header_accum, mi->s_hdr_data, 0))
+               BUG("inbody_header_accum, if not empty, must always contain a valid in-body header");
+       strbuf_reset(&mi->inbody_header_accum);
+}
+
+static int check_inbody_header(struct mailinfo *mi, const struct strbuf *line)
+{
+       if (mi->inbody_header_accum.len &&
+           (line->buf[0] == ' ' || line->buf[0] == '\t')) {
+               if (mi->use_scissors && is_scissors_line(line->buf)) {
+                       /*
+                        * This is a scissors line; do not consider this line
+                        * as a header continuation line.
+                        */
+                       flush_inbody_header_accum(mi);
+                       return 0;
+               }
+               strbuf_strip_suffix(&mi->inbody_header_accum, "\n");
+               strbuf_addbuf(&mi->inbody_header_accum, line);
+               return 1;
+       }
+
+       flush_inbody_header_accum(mi);
+
+       if (starts_with(line->buf, ">From") && isspace(line->buf[5]))
+               return is_format_patch_separator(line->buf + 1, line->len - 1);
+       if (starts_with(line->buf, "[PATCH]") && isspace(line->buf[7])) {
+               int i;
+               for (i = 0; header[i]; i++)
+                       if (!strcmp("Subject", header[i])) {
+                               handle_header(&mi->s_hdr_data[i], line);
+                               return 1;
+                       }
+               return 0;
+       }
+       if (is_inbody_header(mi, line)) {
+               strbuf_addbuf(&mi->inbody_header_accum, line);
+               return 1;
+       }
+       return 0;
+}
+
+static int handle_commit_msg(struct mailinfo *mi, struct strbuf *line)
+{
+       assert(!mi->filter_stage);
+
+       if (mi->header_stage) {
+               if (!line->len || (line->len == 1 && line->buf[0] == '\n')) {
+                       if (mi->inbody_header_accum.len) {
+                               flush_inbody_header_accum(mi);
+                               mi->header_stage = 0;
+                       }
+                       return 0;
+               }
+       }
+
+       if (mi->use_inbody_headers && mi->header_stage) {
+               mi->header_stage = check_inbody_header(mi, line);
+               if (mi->header_stage)
+                       return 0;
+       } else
+               /* Only trim the first (blank) line of the commit message
+                * when ignoring in-body headers.
+                */
+               mi->header_stage = 0;
+
+       /* normalize the log message to UTF-8. */
+       if (convert_to_utf8(mi, line, mi->charset.buf))
+               return 0; /* mi->input_error already set */
+
+       if (mi->use_scissors && is_scissors_line(line->buf)) {
+               int i;
+
+               strbuf_setlen(&mi->log_message, 0);
+               mi->header_stage = 1;
+
+               /*
+                * We may have already read "secondary headers"; purge
+                * them to give ourselves a clean restart.
+                */
+               for (i = 0; header[i]; i++) {
+                       if (mi->s_hdr_data[i])
+                               strbuf_release(mi->s_hdr_data[i]);
+                       mi->s_hdr_data[i] = NULL;
+               }
+               return 0;
+       }
+
+       if (patchbreak(line)) {
+               if (mi->message_id)
+                       strbuf_addf(&mi->log_message,
+                                   "Message-Id: %s\n", mi->message_id);
+               return 1;
+       }
+
+       strbuf_addbuf(&mi->log_message, line);
+       return 0;
+}
+
+static void handle_patch(struct mailinfo *mi, const struct strbuf *line)
+{
+       fwrite(line->buf, 1, line->len, mi->patchfile);
+       mi->patch_lines++;
+}
+
+static void handle_filter(struct mailinfo *mi, struct strbuf *line)
+{
+       switch (mi->filter_stage) {
+       case 0:
+               if (!handle_commit_msg(mi, line))
+                       break;
+               mi->filter_stage++;
+               /* fallthrough */
+       case 1:
+               handle_patch(mi, line);
+               break;
+       }
+}
+
+static int is_rfc2822_header(const struct strbuf *line)
+{
+       /*
+        * The section that defines the loosest possible
+        * field name is "3.6.8 Optional fields".
+        *
+        * optional-field = field-name ":" unstructured CRLF
+        * field-name = 1*ftext
+        * ftext = %d33-57 / %59-126
+        */
+       int ch;
+       char *cp = line->buf;
+
+       /* Count mbox From headers as headers */
+       if (starts_with(cp, "From ") || starts_with(cp, ">From "))
+               return 1;
+
+       while ((ch = *cp++)) {
+               if (ch == ':')
+                       return 1;
+               if ((33 <= ch && ch <= 57) ||
+                   (59 <= ch && ch <= 126))
+                       continue;
+               break;
+       }
+       return 0;
+}
+
+static int read_one_header_line(struct strbuf *line, FILE *in)
+{
+       struct strbuf continuation = STRBUF_INIT;
+
+       /* Get the first part of the line. */
+       if (strbuf_getline_lf(line, in))
+               return 0;
+
+       /*
+        * Is it an empty line or not a valid rfc2822 header?
+        * If so, stop here, and return false ("not a header")
+        */
+       strbuf_rtrim(line);
+       if (!line->len || !is_rfc2822_header(line)) {
+               /* Re-add the newline */
+               strbuf_addch(line, '\n');
+               return 0;
+       }
+
+       /*
+        * Now we need to eat all the continuation lines..
+        * Yuck, 2822 header "folding"
+        */
+       for (;;) {
+               int peek;
+
+               peek = fgetc(in);
+               if (peek == EOF)
+                       break;
+               ungetc(peek, in);
+               if (peek != ' ' && peek != '\t')
+                       break;
+               if (strbuf_getline_lf(&continuation, in))
+                       break;
+               continuation.buf[0] = ' ';
+               strbuf_rtrim(&continuation);
+               strbuf_addbuf(line, &continuation);
+       }
+       strbuf_release(&continuation);
+
+       return 1;
+}
+
+static int find_boundary(struct mailinfo *mi, struct strbuf *line)
+{
+       while (!strbuf_getline_lf(line, mi->input)) {
+               if (*(mi->content_top) && is_multipart_boundary(mi, line))
+                       return 1;
+       }
+       return 0;
+}
+
+static int handle_boundary(struct mailinfo *mi, struct strbuf *line)
+{
+       struct strbuf newline = STRBUF_INIT;
+
+       strbuf_addch(&newline, '\n');
+again:
+       if (line->len >= (*(mi->content_top))->len + 2 &&
+           !memcmp(line->buf + (*(mi->content_top))->len, "--", 2)) {
+               /* we hit an end boundary */
+               /* pop the current boundary off the stack */
+               strbuf_release(*(mi->content_top));
+               FREE_AND_NULL(*(mi->content_top));
+
+               /* technically won't happen as is_multipart_boundary()
+                  will fail first.  But just in case..
+                */
+               if (--mi->content_top < mi->content) {
+                       error("Detected mismatched boundaries, can't recover");
+                       mi->input_error = -1;
+                       mi->content_top = mi->content;
+                       strbuf_release(&newline);
+                       return 0;
+               }
+               handle_filter(mi, &newline);
+               strbuf_release(&newline);
+               if (mi->input_error)
+                       return 0;
+
+               /* skip to the next boundary */
+               if (!find_boundary(mi, line))
+                       return 0;
+               goto again;
+       }
+
+       /* set some defaults */
+       mi->transfer_encoding = TE_DONTCARE;
+       strbuf_reset(&mi->charset);
+
+       /* slurp in this section's info */
+       while (read_one_header_line(line, mi->input))
+               check_header(mi, line, mi->p_hdr_data, 0);
+
+       strbuf_release(&newline);
+       /* replenish line */
+       if (strbuf_getline_lf(line, mi->input))
+               return 0;
+       strbuf_addch(line, '\n');
+       return 1;
+}
+
+static void handle_filter_flowed(struct mailinfo *mi, struct strbuf *line,
+                                struct strbuf *prev)
+{
+       size_t len = line->len;
+       const char *rest;
+
+       if (!mi->format_flowed) {
+               handle_filter(mi, line);
+               return;
+       }
+
+       if (line->buf[len - 1] == '\n') {
+               len--;
+               if (len && line->buf[len - 1] == '\r')
+                       len--;
+       }
+
+       /* Keep signature separator as-is. */
+       if (skip_prefix(line->buf, "-- ", &rest) && rest - line->buf == len) {
+               if (prev->len) {
+                       handle_filter(mi, prev);
+                       strbuf_reset(prev);
+               }
+               handle_filter(mi, line);
+               return;
+       }
+
+       /* Unstuff space-stuffed line. */
+       if (len && line->buf[0] == ' ') {
+               strbuf_remove(line, 0, 1);
+               len--;
+       }
+
+       /* Save flowed line for later, but without the soft line break. */
+       if (len && line->buf[len - 1] == ' ') {
+               strbuf_add(prev, line->buf, len - !!mi->delsp);
+               return;
+       }
+
+       /* Prepend any previous partial lines */
+       strbuf_insert(line, 0, prev->buf, prev->len);
+       strbuf_reset(prev);
+
+       handle_filter(mi, line);
+}
+
+static void handle_body(struct mailinfo *mi, struct strbuf *line)
+{
+       struct strbuf prev = STRBUF_INIT;
+
+       /* Skip up to the first boundary */
+       if (*(mi->content_top)) {
+               if (!find_boundary(mi, line))
+                       goto handle_body_out;
+       }
+
+       do {
+               /* process any boundary lines */
+               if (*(mi->content_top) && is_multipart_boundary(mi, line)) {
+                       /* flush any leftover */
+                       if (prev.len) {
+                               handle_filter(mi, &prev);
+                               strbuf_reset(&prev);
+                       }
+                       if (!handle_boundary(mi, line))
+                               goto handle_body_out;
+               }
+
+               /* Unwrap transfer encoding */
+               decode_transfer_encoding(mi, line);
+
+               switch (mi->transfer_encoding) {
+               case TE_BASE64:
+               case TE_QP:
+               {
+                       struct strbuf **lines, **it, *sb;
+
+                       /* Prepend any previous partial lines */
+                       strbuf_insert(line, 0, prev.buf, prev.len);
+                       strbuf_reset(&prev);
+
+                       /*
+                        * This is a decoded line that may contain
+                        * multiple new lines.  Pass only one chunk
+                        * at a time to handle_filter()
+                        */
+                       lines = strbuf_split(line, '\n');
+                       for (it = lines; (sb = *it); it++) {
+                               if (*(it + 1) == NULL) /* The last line */
+                                       if (sb->buf[sb->len - 1] != '\n') {
+                                               /* Partial line, save it for later. */
+                                               strbuf_addbuf(&prev, sb);
+                                               break;
+                                       }
+                               handle_filter_flowed(mi, sb, &prev);
+                       }
+                       /*
+                        * The partial chunk is saved in "prev" and will be
+                        * appended by the next iteration of read_line_with_nul().
+                        */
+                       strbuf_list_free(lines);
+                       break;
+               }
+               default:
+                       handle_filter_flowed(mi, line, &prev);
+               }
+
+               if (mi->input_error)
+                       break;
+       } while (!strbuf_getwholeline(line, mi->input, '\n'));
+
+       if (prev.len)
+               handle_filter(mi, &prev);
+
+       flush_inbody_header_accum(mi);
+
+handle_body_out:
+       strbuf_release(&prev);
+}
+
+static void output_header_lines(FILE *fout, const char *hdr, const struct strbuf *data)
+{
+       const char *sp = data->buf;
+       while (1) {
+               char *ep = strchr(sp, '\n');
+               int len;
+               if (!ep)
+                       len = strlen(sp);
+               else
+                       len = ep - sp;
+               fprintf(fout, "%s: %.*s\n", hdr, len, sp);
+               if (!ep)
+                       break;
+               sp = ep + 1;
+       }
+}
+
+static void handle_info(struct mailinfo *mi)
+{
+       struct strbuf *hdr;
+       int i;
+
+       for (i = 0; header[i]; i++) {
+               /* only print inbody headers if we output a patch file */
+               if (mi->patch_lines && mi->s_hdr_data[i])
+                       hdr = mi->s_hdr_data[i];
+               else if (mi->p_hdr_data[i])
+                       hdr = mi->p_hdr_data[i];
+               else
+                       continue;
+
+               if (!strcmp(header[i], "Subject")) {
+                       if (!mi->keep_subject) {
+                               cleanup_subject(mi, hdr);
+                               cleanup_space(hdr);
+                       }
+                       output_header_lines(mi->output, "Subject", hdr);
+               } else if (!strcmp(header[i], "From")) {
+                       cleanup_space(hdr);
+                       handle_from(mi, hdr);
+                       fprintf(mi->output, "Author: %s\n", mi->name.buf);
+                       fprintf(mi->output, "Email: %s\n", mi->email.buf);
+               } else {
+                       cleanup_space(hdr);
+                       fprintf(mi->output, "%s: %s\n", header[i], hdr->buf);
+               }
+       }
+       fprintf(mi->output, "\n");
+}
+
+int mailinfo(struct mailinfo *mi, const char *msg, const char *patch)
+{
+       FILE *cmitmsg;
+       int peek;
+       struct strbuf line = STRBUF_INIT;
+
+       cmitmsg = fopen(msg, "w");
+       if (!cmitmsg) {
+               perror(msg);
+               return -1;
+       }
+       mi->patchfile = fopen(patch, "w");
+       if (!mi->patchfile) {
+               perror(patch);
+               fclose(cmitmsg);
+               return -1;
+       }
+
+       mi->p_hdr_data = xcalloc(MAX_HDR_PARSED, sizeof(*(mi->p_hdr_data)));
+       mi->s_hdr_data = xcalloc(MAX_HDR_PARSED, sizeof(*(mi->s_hdr_data)));
+
+       do {
+               peek = fgetc(mi->input);
+               if (peek == EOF) {
+                       fclose(cmitmsg);
+                       return error("empty patch: '%s'", patch);
+               }
+       } while (isspace(peek));
+       ungetc(peek, mi->input);
+
+       /* process the email header */
+       while (read_one_header_line(&line, mi->input))
+               check_header(mi, &line, mi->p_hdr_data, 1);
+
+       handle_body(mi, &line);
+       fwrite(mi->log_message.buf, 1, mi->log_message.len, cmitmsg);
+       fclose(cmitmsg);
+       fclose(mi->patchfile);
+
+       handle_info(mi);
+       strbuf_release(&line);
+       return mi->input_error;
+}
+
+static int git_mailinfo_config(const char *var, const char *value, void *mi_)
+{
+       struct mailinfo *mi = mi_;
+
+       if (!starts_with(var, "mailinfo."))
+               return git_default_config(var, value, NULL);
+       if (!strcmp(var, "mailinfo.scissors")) {
+               mi->use_scissors = git_config_bool(var, value);
+               return 0;
+       }
+       /* perhaps others here */
+       return 0;
+}
+
+void setup_mailinfo(struct mailinfo *mi)
+{
+       memset(mi, 0, sizeof(*mi));
+       strbuf_init(&mi->name, 0);
+       strbuf_init(&mi->email, 0);
+       strbuf_init(&mi->charset, 0);
+       strbuf_init(&mi->log_message, 0);
+       strbuf_init(&mi->inbody_header_accum, 0);
+       mi->header_stage = 1;
+       mi->use_inbody_headers = 1;
+       mi->content_top = mi->content;
+       git_config(git_mailinfo_config, mi);
+}
+
+void clear_mailinfo(struct mailinfo *mi)
+{
+       int i;
+
+       strbuf_release(&mi->name);
+       strbuf_release(&mi->email);
+       strbuf_release(&mi->charset);
+       strbuf_release(&mi->inbody_header_accum);
+       free(mi->message_id);
+
+       if (mi->p_hdr_data)
+               for (i = 0; mi->p_hdr_data[i]; i++)
+                       strbuf_release(mi->p_hdr_data[i]);
+       free(mi->p_hdr_data);
+       if (mi->s_hdr_data)
+               for (i = 0; mi->s_hdr_data[i]; i++)
+                       strbuf_release(mi->s_hdr_data[i]);
+       free(mi->s_hdr_data);
+
+       while (mi->content < mi->content_top) {
+               free(*(mi->content_top));
+               mi->content_top--;
+       }
+
+       strbuf_release(&mi->log_message);
+}
diff --git a/t/t4256/1/mailinfo.c.orig b/t/t4256/1/mailinfo.c.orig
new file mode 100644 (file)
index 0000000..3281a37
--- /dev/null
@@ -0,0 +1,1185 @@
+#include "cache.h"
+#include "config.h"
+#include "utf8.h"
+#include "strbuf.h"
+#include "mailinfo.h"
+
+static void cleanup_space(struct strbuf *sb)
+{
+       size_t pos, cnt;
+       for (pos = 0; pos < sb->len; pos++) {
+               if (isspace(sb->buf[pos])) {
+                       sb->buf[pos] = ' ';
+                       for (cnt = 0; isspace(sb->buf[pos + cnt + 1]); cnt++);
+                       strbuf_remove(sb, pos + 1, cnt);
+               }
+       }
+}
+
+static void get_sane_name(struct strbuf *out, struct strbuf *name, struct strbuf *email)
+{
+       struct strbuf *src = name;
+       if (name->len < 3 || 60 < name->len || strchr(name->buf, '@') ||
+               strchr(name->buf, '<') || strchr(name->buf, '>'))
+               src = email;
+       else if (name == out)
+               return;
+       strbuf_reset(out);
+       strbuf_addbuf(out, src);
+}
+
+static void parse_bogus_from(struct mailinfo *mi, const struct strbuf *line)
+{
+       /* John Doe <johndoe> */
+
+       char *bra, *ket;
+       /* This is fallback, so do not bother if we already have an
+        * e-mail address.
+        */
+       if (mi->email.len)
+               return;
+
+       bra = strchr(line->buf, '<');
+       if (!bra)
+               return;
+       ket = strchr(bra, '>');
+       if (!ket)
+               return;
+
+       strbuf_reset(&mi->email);
+       strbuf_add(&mi->email, bra + 1, ket - bra - 1);
+
+       strbuf_reset(&mi->name);
+       strbuf_add(&mi->name, line->buf, bra - line->buf);
+       strbuf_trim(&mi->name);
+       get_sane_name(&mi->name, &mi->name, &mi->email);
+}
+
+static const char *unquote_comment(struct strbuf *outbuf, const char *in)
+{
+       int c;
+       int take_next_literally = 0;
+
+       strbuf_addch(outbuf, '(');
+
+       while ((c = *in++) != 0) {
+               if (take_next_literally == 1) {
+                       take_next_literally = 0;
+               } else {
+                       switch (c) {
+                       case '\\':
+                               take_next_literally = 1;
+                               continue;
+                       case '(':
+                               in = unquote_comment(outbuf, in);
+                               continue;
+                       case ')':
+                               strbuf_addch(outbuf, ')');
+                               return in;
+                       }
+               }
+
+               strbuf_addch(outbuf, c);
+       }
+
+       return in;
+}
+
+static const char *unquote_quoted_string(struct strbuf *outbuf, const char *in)
+{
+       int c;
+       int take_next_literally = 0;
+
+       while ((c = *in++) != 0) {
+               if (take_next_literally == 1) {
+                       take_next_literally = 0;
+               } else {
+                       switch (c) {
+                       case '\\':
+                               take_next_literally = 1;
+                               continue;
+                       case '"':
+                               return in;
+                       }
+               }
+
+               strbuf_addch(outbuf, c);
+       }
+
+       return in;
+}
+
+static void unquote_quoted_pair(struct strbuf *line)
+{
+       struct strbuf outbuf;
+       const char *in = line->buf;
+       int c;
+
+       strbuf_init(&outbuf, line->len);
+
+       while ((c = *in++) != 0) {
+               switch (c) {
+               case '"':
+                       in = unquote_quoted_string(&outbuf, in);
+                       continue;
+               case '(':
+                       in = unquote_comment(&outbuf, in);
+                       continue;
+               }
+
+               strbuf_addch(&outbuf, c);
+       }
+
+       strbuf_swap(&outbuf, line);
+       strbuf_release(&outbuf);
+
+}
+
+static void handle_from(struct mailinfo *mi, const struct strbuf *from)
+{
+       char *at;
+       size_t el;
+       struct strbuf f;
+
+       strbuf_init(&f, from->len);
+       strbuf_addbuf(&f, from);
+
+       unquote_quoted_pair(&f);
+
+       at = strchr(f.buf, '@');
+       if (!at) {
+               parse_bogus_from(mi, from);
+               goto out;
+       }
+
+       /*
+        * If we already have one email, don't take any confusing lines
+        */
+       if (mi->email.len && strchr(at + 1, '@'))
+               goto out;
+
+       /* Pick up the string around '@', possibly delimited with <>
+        * pair; that is the email part.
+        */
+       while (at > f.buf) {
+               char c = at[-1];
+               if (isspace(c))
+                       break;
+               if (c == '<') {
+                       at[-1] = ' ';
+                       break;
+               }
+               at--;
+       }
+       el = strcspn(at, " \n\t\r\v\f>");
+       strbuf_reset(&mi->email);
+       strbuf_add(&mi->email, at, el);
+       strbuf_remove(&f, at - f.buf, el + (at[el] ? 1 : 0));
+
+       /* The remainder is name.  It could be
+        *
+        * - "John Doe <john.doe@xz>"                   (a), or
+        * - "john.doe@xz (John Doe)"                   (b), or
+        * - "John (zzz) Doe <john.doe@xz> (Comment)"   (c)
+        *
+        * but we have removed the email part, so
+        *
+        * - remove extra spaces which could stay after email (case 'c'), and
+        * - trim from both ends, possibly removing the () pair at the end
+        *   (cases 'a' and 'b').
+        */
+       cleanup_space(&f);
+       strbuf_trim(&f);
+       if (f.buf[0] == '(' && f.len && f.buf[f.len - 1] == ')') {
+               strbuf_remove(&f, 0, 1);
+               strbuf_setlen(&f, f.len - 1);
+       }
+
+       get_sane_name(&mi->name, &f, &mi->email);
+out:
+       strbuf_release(&f);
+}
+
+static void handle_header(struct strbuf **out, const struct strbuf *line)
+{
+       if (!*out) {
+               *out = xmalloc(sizeof(struct strbuf));
+               strbuf_init(*out, line->len);
+       } else
+               strbuf_reset(*out);
+
+       strbuf_addbuf(*out, line);
+}
+
+/* NOTE NOTE NOTE.  We do not claim we do full MIME.  We just attempt
+ * to have enough heuristics to grok MIME encoded patches often found
+ * on our mailing lists.  For example, we do not even treat header lines
+ * case insensitively.
+ */
+
+static int slurp_attr(const char *line, const char *name, struct strbuf *attr)
+{
+       const char *ends, *ap = strcasestr(line, name);
+       size_t sz;
+
+       strbuf_setlen(attr, 0);
+       if (!ap)
+               return 0;
+       ap += strlen(name);
+       if (*ap == '"') {
+               ap++;
+               ends = "\"";
+       }
+       else
+               ends = "; \t";
+       sz = strcspn(ap, ends);
+       strbuf_add(attr, ap, sz);
+       return 1;
+}
+
+static void handle_content_type(struct mailinfo *mi, struct strbuf *line)
+{
+       struct strbuf *boundary = xmalloc(sizeof(struct strbuf));
+       strbuf_init(boundary, line->len);
+
+       if (slurp_attr(line->buf, "boundary=", boundary)) {
+               strbuf_insert(boundary, 0, "--", 2);
+               if (++mi->content_top >= &mi->content[MAX_BOUNDARIES]) {
+                       error("Too many boundaries to handle");
+                       mi->input_error = -1;
+                       mi->content_top = &mi->content[MAX_BOUNDARIES] - 1;
+                       return;
+               }
+               *(mi->content_top) = boundary;
+               boundary = NULL;
+       }
+       slurp_attr(line->buf, "charset=", &mi->charset);
+
+       if (boundary) {
+               strbuf_release(boundary);
+               free(boundary);
+       }
+}
+
+static void handle_content_transfer_encoding(struct mailinfo *mi,
+                                            const struct strbuf *line)
+{
+       if (strcasestr(line->buf, "base64"))
+               mi->transfer_encoding = TE_BASE64;
+       else if (strcasestr(line->buf, "quoted-printable"))
+               mi->transfer_encoding = TE_QP;
+       else
+               mi->transfer_encoding = TE_DONTCARE;
+}
+
+static int is_multipart_boundary(struct mailinfo *mi, const struct strbuf *line)
+{
+       struct strbuf *content_top = *(mi->content_top);
+
+       return ((content_top->len <= line->len) &&
+               !memcmp(line->buf, content_top->buf, content_top->len));
+}
+
+static void cleanup_subject(struct mailinfo *mi, struct strbuf *subject)
+{
+       size_t at = 0;
+
+       while (at < subject->len) {
+               char *pos;
+               size_t remove;
+
+               switch (subject->buf[at]) {
+               case 'r': case 'R':
+                       if (subject->len <= at + 3)
+                               break;
+                       if ((subject->buf[at + 1] == 'e' ||
+                            subject->buf[at + 1] == 'E') &&
+                           subject->buf[at + 2] == ':') {
+                               strbuf_remove(subject, at, 3);
+                               continue;
+                       }
+                       at++;
+                       break;
+               case ' ': case '\t': case ':':
+                       strbuf_remove(subject, at, 1);
+                       continue;
+               case '[':
+                       pos = strchr(subject->buf + at, ']');
+                       if (!pos)
+                               break;
+                       remove = pos - subject->buf + at + 1;
+                       if (!mi->keep_non_patch_brackets_in_subject ||
+                           (7 <= remove &&
+                            memmem(subject->buf + at, remove, "PATCH", 5)))
+                               strbuf_remove(subject, at, remove);
+                       else {
+                               at += remove;
+                               /*
+                                * If the input had a space after the ], keep
+                                * it.  We don't bother with finding the end of
+                                * the space, since we later normalize it
+                                * anyway.
+                                */
+                               if (isspace(subject->buf[at]))
+                                       at += 1;
+                       }
+                       continue;
+               }
+               break;
+       }
+       strbuf_trim(subject);
+}
+
+#define MAX_HDR_PARSED 10
+static const char *header[MAX_HDR_PARSED] = {
+       "From","Subject","Date",
+};
+
+static inline int cmp_header(const struct strbuf *line, const char *hdr)
+{
+       int len = strlen(hdr);
+       return !strncasecmp(line->buf, hdr, len) && line->len > len &&
+                       line->buf[len] == ':' && isspace(line->buf[len + 1]);
+}
+
+static int is_format_patch_separator(const char *line, int len)
+{
+       static const char SAMPLE[] =
+               "From e6807f3efca28b30decfecb1732a56c7db1137ee Mon Sep 17 00:00:00 2001\n";
+       const char *cp;
+
+       if (len != strlen(SAMPLE))
+               return 0;
+       if (!skip_prefix(line, "From ", &cp))
+               return 0;
+       if (strspn(cp, "0123456789abcdef") != 40)
+               return 0;
+       cp += 40;
+       return !memcmp(SAMPLE + (cp - line), cp, strlen(SAMPLE) - (cp - line));
+}
+
+static struct strbuf *decode_q_segment(const struct strbuf *q_seg, int rfc2047)
+{
+       const char *in = q_seg->buf;
+       int c;
+       struct strbuf *out = xmalloc(sizeof(struct strbuf));
+       strbuf_init(out, q_seg->len);
+
+       while ((c = *in++) != 0) {
+               if (c == '=') {
+                       int ch, d = *in;
+                       if (d == '\n' || !d)
+                               break; /* drop trailing newline */
+                       ch = hex2chr(in);
+                       if (ch >= 0) {
+                               strbuf_addch(out, ch);
+                               in += 2;
+                               continue;
+                       }
+                       /* garbage -- fall through */
+               }
+               if (rfc2047 && c == '_') /* rfc2047 4.2 (2) */
+                       c = 0x20;
+               strbuf_addch(out, c);
+       }
+       return out;
+}
+
+static struct strbuf *decode_b_segment(const struct strbuf *b_seg)
+{
+       /* Decode in..ep, possibly in-place to ot */
+       int c, pos = 0, acc = 0;
+       const char *in = b_seg->buf;
+       struct strbuf *out = xmalloc(sizeof(struct strbuf));
+       strbuf_init(out, b_seg->len);
+
+       while ((c = *in++) != 0) {
+               if (c == '+')
+                       c = 62;
+               else if (c == '/')
+                       c = 63;
+               else if ('A' <= c && c <= 'Z')
+                       c -= 'A';
+               else if ('a' <= c && c <= 'z')
+                       c -= 'a' - 26;
+               else if ('0' <= c && c <= '9')
+                       c -= '0' - 52;
+               else
+                       continue; /* garbage */
+               switch (pos++) {
+               case 0:
+                       acc = (c << 2);
+                       break;
+               case 1:
+                       strbuf_addch(out, (acc | (c >> 4)));
+                       acc = (c & 15) << 4;
+                       break;
+               case 2:
+                       strbuf_addch(out, (acc | (c >> 2)));
+                       acc = (c & 3) << 6;
+                       break;
+               case 3:
+                       strbuf_addch(out, (acc | c));
+                       acc = pos = 0;
+                       break;
+               }
+       }
+       return out;
+}
+
+static int convert_to_utf8(struct mailinfo *mi,
+                          struct strbuf *line, const char *charset)
+{
+       char *out;
+
+       if (!mi->metainfo_charset || !charset || !*charset)
+               return 0;
+
+       if (same_encoding(mi->metainfo_charset, charset))
+               return 0;
+       out = reencode_string(line->buf, mi->metainfo_charset, charset);
+       if (!out) {
+               mi->input_error = -1;
+               return error("cannot convert from %s to %s",
+                            charset, mi->metainfo_charset);
+       }
+       strbuf_attach(line, out, strlen(out), strlen(out));
+       return 0;
+}
+
+static void decode_header(struct mailinfo *mi, struct strbuf *it)
+{
+       char *in, *ep, *cp;
+       struct strbuf outbuf = STRBUF_INIT, *dec;
+       struct strbuf charset_q = STRBUF_INIT, piecebuf = STRBUF_INIT;
+       int found_error = 1; /* pessimism */
+
+       in = it->buf;
+       while (in - it->buf <= it->len && (ep = strstr(in, "=?")) != NULL) {
+               int encoding;
+               strbuf_reset(&charset_q);
+               strbuf_reset(&piecebuf);
+
+               if (in != ep) {
+                       /*
+                        * We are about to process an encoded-word
+                        * that begins at ep, but there is something
+                        * before the encoded word.
+                        */
+                       char *scan;
+                       for (scan = in; scan < ep; scan++)
+                               if (!isspace(*scan))
+                                       break;
+
+                       if (scan != ep || in == it->buf) {
+                               /*
+                                * We should not lose that "something",
+                                * unless we have just processed an
+                                * encoded-word, and there is only LWS
+                                * before the one we are about to process.
+                                */
+                               strbuf_add(&outbuf, in, ep - in);
+                       }
+               }
+               /* E.g.
+                * ep : "=?iso-2022-jp?B?GyR...?= foo"
+                * ep : "=?ISO-8859-1?Q?Foo=FCbar?= baz"
+                */
+               ep += 2;
+
+               if (ep - it->buf >= it->len || !(cp = strchr(ep, '?')))
+                       goto release_return;
+
+               if (cp + 3 - it->buf > it->len)
+                       goto release_return;
+               strbuf_add(&charset_q, ep, cp - ep);
+
+               encoding = cp[1];
+               if (!encoding || cp[2] != '?')
+                       goto release_return;
+               ep = strstr(cp + 3, "?=");
+               if (!ep)
+                       goto release_return;
+               strbuf_add(&piecebuf, cp + 3, ep - cp - 3);
+               switch (tolower(encoding)) {
+               default:
+                       goto release_return;
+               case 'b':
+                       dec = decode_b_segment(&piecebuf);
+                       break;
+               case 'q':
+                       dec = decode_q_segment(&piecebuf, 1);
+                       break;
+               }
+               if (convert_to_utf8(mi, dec, charset_q.buf))
+                       goto release_return;
+
+               strbuf_addbuf(&outbuf, dec);
+               strbuf_release(dec);
+               free(dec);
+               in = ep + 2;
+       }
+       strbuf_addstr(&outbuf, in);
+       strbuf_reset(it);
+       strbuf_addbuf(it, &outbuf);
+       found_error = 0;
+release_return:
+       strbuf_release(&outbuf);
+       strbuf_release(&charset_q);
+       strbuf_release(&piecebuf);
+
+       if (found_error)
+               mi->input_error = -1;
+}
+
+static int check_header(struct mailinfo *mi,
+                       const struct strbuf *line,
+                       struct strbuf *hdr_data[], int overwrite)
+{
+       int i, ret = 0, len;
+       struct strbuf sb = STRBUF_INIT;
+
+       /* search for the interesting parts */
+       for (i = 0; header[i]; i++) {
+               int len = strlen(header[i]);
+               if ((!hdr_data[i] || overwrite) && cmp_header(line, header[i])) {
+                       /* Unwrap inline B and Q encoding, and optionally
+                        * normalize the meta information to utf8.
+                        */
+                       strbuf_add(&sb, line->buf + len + 2, line->len - len - 2);
+                       decode_header(mi, &sb);
+                       handle_header(&hdr_data[i], &sb);
+                       ret = 1;
+                       goto check_header_out;
+               }
+       }
+
+       /* Content stuff */
+       if (cmp_header(line, "Content-Type")) {
+               len = strlen("Content-Type: ");
+               strbuf_add(&sb, line->buf + len, line->len - len);
+               decode_header(mi, &sb);
+               strbuf_insert(&sb, 0, "Content-Type: ", len);
+               handle_content_type(mi, &sb);
+               ret = 1;
+               goto check_header_out;
+       }
+       if (cmp_header(line, "Content-Transfer-Encoding")) {
+               len = strlen("Content-Transfer-Encoding: ");
+               strbuf_add(&sb, line->buf + len, line->len - len);
+               decode_header(mi, &sb);
+               handle_content_transfer_encoding(mi, &sb);
+               ret = 1;
+               goto check_header_out;
+       }
+       if (cmp_header(line, "Message-Id")) {
+               len = strlen("Message-Id: ");
+               strbuf_add(&sb, line->buf + len, line->len - len);
+               decode_header(mi, &sb);
+               if (mi->add_message_id)
+                       mi->message_id = strbuf_detach(&sb, NULL);
+               ret = 1;
+               goto check_header_out;
+       }
+
+check_header_out:
+       strbuf_release(&sb);
+       return ret;
+}
+
+/*
+ * Returns 1 if the given line or any line beginning with the given line is an
+ * in-body header (that is, check_header will succeed when passed
+ * mi->s_hdr_data).
+ */
+static int is_inbody_header(const struct mailinfo *mi,
+                           const struct strbuf *line)
+{
+       int i;
+       for (i = 0; header[i]; i++)
+               if (!mi->s_hdr_data[i] && cmp_header(line, header[i]))
+                       return 1;
+       return 0;
+}
+
+static void decode_transfer_encoding(struct mailinfo *mi, struct strbuf *line)
+{
+       struct strbuf *ret;
+
+       switch (mi->transfer_encoding) {
+       case TE_QP:
+               ret = decode_q_segment(line, 0);
+               break;
+       case TE_BASE64:
+               ret = decode_b_segment(line);
+               break;
+       case TE_DONTCARE:
+       default:
+               return;
+       }
+       strbuf_reset(line);
+       strbuf_addbuf(line, ret);
+       strbuf_release(ret);
+       free(ret);
+}
+
+static inline int patchbreak(const struct strbuf *line)
+{
+       size_t i;
+
+       /* Beginning of a "diff -" header? */
+       if (starts_with(line->buf, "diff -"))
+               return 1;
+
+       /* CVS "Index: " line? */
+       if (starts_with(line->buf, "Index: "))
+               return 1;
+
+       /*
+        * "--- <filename>" starts patches without headers
+        * "---<sp>*" is a manual separator
+        */
+       if (line->len < 4)
+               return 0;
+
+       if (starts_with(line->buf, "---")) {
+               /* space followed by a filename? */
+               if (line->buf[3] == ' ' && !isspace(line->buf[4]))
+                       return 1;
+               /* Just whitespace? */
+               for (i = 3; i < line->len; i++) {
+                       unsigned char c = line->buf[i];
+                       if (c == '\n')
+                               return 1;
+                       if (!isspace(c))
+                               break;
+               }
+               return 0;
+       }
+       return 0;
+}
+
+static int is_scissors_line(const char *line)
+{
+       const char *c;
+       int scissors = 0, gap = 0;
+       const char *first_nonblank = NULL, *last_nonblank = NULL;
+       int visible, perforation = 0, in_perforation = 0;
+
+       for (c = line; *c; c++) {
+               if (isspace(*c)) {
+                       if (in_perforation) {
+                               perforation++;
+                               gap++;
+                       }
+                       continue;
+               }
+               last_nonblank = c;
+               if (first_nonblank == NULL)
+                       first_nonblank = c;
+               if (*c == '-') {
+                       in_perforation = 1;
+                       perforation++;
+                       continue;
+               }
+               if ((!memcmp(c, ">8", 2) || !memcmp(c, "8<", 2) ||
+                    !memcmp(c, ">%", 2) || !memcmp(c, "%<", 2))) {
+                       in_perforation = 1;
+                       perforation += 2;
+                       scissors += 2;
+                       c++;
+                       continue;
+               }
+               in_perforation = 0;
+       }
+
+       /*
+        * The mark must be at least 8 bytes long (e.g. "-- >8 --").
+        * Even though there can be arbitrary cruft on the same line
+        * (e.g. "cut here"), in order to avoid misidentification, the
+        * perforation must occupy more than a third of the visible
+        * width of the line, and dashes and scissors must occupy more
+        * than half of the perforation.
+        */
+
+       if (first_nonblank && last_nonblank)
+               visible = last_nonblank - first_nonblank + 1;
+       else
+               visible = 0;
+       return (scissors && 8 <= visible &&
+               visible < perforation * 3 &&
+               gap * 2 < perforation);
+}
+
+static void flush_inbody_header_accum(struct mailinfo *mi)
+{
+       if (!mi->inbody_header_accum.len)
+               return;
+       if (!check_header(mi, &mi->inbody_header_accum, mi->s_hdr_data, 0))
+               BUG("inbody_header_accum, if not empty, must always contain a valid in-body header");
+       strbuf_reset(&mi->inbody_header_accum);
+}
+
+static int check_inbody_header(struct mailinfo *mi, const struct strbuf *line)
+{
+       if (mi->inbody_header_accum.len &&
+           (line->buf[0] == ' ' || line->buf[0] == '\t')) {
+               if (mi->use_scissors && is_scissors_line(line->buf)) {
+                       /*
+                        * This is a scissors line; do not consider this line
+                        * as a header continuation line.
+                        */
+                       flush_inbody_header_accum(mi);
+                       return 0;
+               }
+               strbuf_strip_suffix(&mi->inbody_header_accum, "\n");
+               strbuf_addbuf(&mi->inbody_header_accum, line);
+               return 1;
+       }
+
+       flush_inbody_header_accum(mi);
+
+       if (starts_with(line->buf, ">From") && isspace(line->buf[5]))
+               return is_format_patch_separator(line->buf + 1, line->len - 1);
+       if (starts_with(line->buf, "[PATCH]") && isspace(line->buf[7])) {
+               int i;
+               for (i = 0; header[i]; i++)
+                       if (!strcmp("Subject", header[i])) {
+                               handle_header(&mi->s_hdr_data[i], line);
+                               return 1;
+                       }
+               return 0;
+       }
+       if (is_inbody_header(mi, line)) {
+               strbuf_addbuf(&mi->inbody_header_accum, line);
+               return 1;
+       }
+       return 0;
+}
+
+static int handle_commit_msg(struct mailinfo *mi, struct strbuf *line)
+{
+       assert(!mi->filter_stage);
+
+       if (mi->header_stage) {
+               if (!line->len || (line->len == 1 && line->buf[0] == '\n')) {
+                       if (mi->inbody_header_accum.len) {
+                               flush_inbody_header_accum(mi);
+                               mi->header_stage = 0;
+                       }
+                       return 0;
+               }
+       }
+
+       if (mi->use_inbody_headers && mi->header_stage) {
+               mi->header_stage = check_inbody_header(mi, line);
+               if (mi->header_stage)
+                       return 0;
+       } else
+               /* Only trim the first (blank) line of the commit message
+                * when ignoring in-body headers.
+                */
+               mi->header_stage = 0;
+
+       /* normalize the log message to UTF-8. */
+       if (convert_to_utf8(mi, line, mi->charset.buf))
+               return 0; /* mi->input_error already set */
+
+       if (mi->use_scissors && is_scissors_line(line->buf)) {
+               int i;
+
+               strbuf_setlen(&mi->log_message, 0);
+               mi->header_stage = 1;
+
+               /*
+                * We may have already read "secondary headers"; purge
+                * them to give ourselves a clean restart.
+                */
+               for (i = 0; header[i]; i++) {
+                       if (mi->s_hdr_data[i])
+                               strbuf_release(mi->s_hdr_data[i]);
+                       mi->s_hdr_data[i] = NULL;
+               }
+               return 0;
+       }
+
+       if (patchbreak(line)) {
+               if (mi->message_id)
+                       strbuf_addf(&mi->log_message,
+                                   "Message-Id: %s\n", mi->message_id);
+               return 1;
+       }
+
+       strbuf_addbuf(&mi->log_message, line);
+       return 0;
+}
+
+static void handle_patch(struct mailinfo *mi, const struct strbuf *line)
+{
+       fwrite(line->buf, 1, line->len, mi->patchfile);
+       mi->patch_lines++;
+}
+
+static void handle_filter(struct mailinfo *mi, struct strbuf *line)
+{
+       switch (mi->filter_stage) {
+       case 0:
+               if (!handle_commit_msg(mi, line))
+                       break;
+               mi->filter_stage++;
+               /* fallthrough */
+       case 1:
+               handle_patch(mi, line);
+               break;
+       }
+}
+
+static int is_rfc2822_header(const struct strbuf *line)
+{
+       /*
+        * The section that defines the loosest possible
+        * field name is "3.6.8 Optional fields".
+        *
+        * optional-field = field-name ":" unstructured CRLF
+        * field-name = 1*ftext
+        * ftext = %d33-57 / %59-126
+        */
+       int ch;
+       char *cp = line->buf;
+
+       /* Count mbox From headers as headers */
+       if (starts_with(cp, "From ") || starts_with(cp, ">From "))
+               return 1;
+
+       while ((ch = *cp++)) {
+               if (ch == ':')
+                       return 1;
+               if ((33 <= ch && ch <= 57) ||
+                   (59 <= ch && ch <= 126))
+                       continue;
+               break;
+       }
+       return 0;
+}
+
+static int read_one_header_line(struct strbuf *line, FILE *in)
+{
+       struct strbuf continuation = STRBUF_INIT;
+
+       /* Get the first part of the line. */
+       if (strbuf_getline_lf(line, in))
+               return 0;
+
+       /*
+        * Is it an empty line or not a valid rfc2822 header?
+        * If so, stop here, and return false ("not a header")
+        */
+       strbuf_rtrim(line);
+       if (!line->len || !is_rfc2822_header(line)) {
+               /* Re-add the newline */
+               strbuf_addch(line, '\n');
+               return 0;
+       }
+
+       /*
+        * Now we need to eat all the continuation lines..
+        * Yuck, 2822 header "folding"
+        */
+       for (;;) {
+               int peek;
+
+               peek = fgetc(in);
+               if (peek == EOF)
+                       break;
+               ungetc(peek, in);
+               if (peek != ' ' && peek != '\t')
+                       break;
+               if (strbuf_getline_lf(&continuation, in))
+                       break;
+               continuation.buf[0] = ' ';
+               strbuf_rtrim(&continuation);
+               strbuf_addbuf(line, &continuation);
+       }
+       strbuf_release(&continuation);
+
+       return 1;
+}
+
+static int find_boundary(struct mailinfo *mi, struct strbuf *line)
+{
+       while (!strbuf_getline_lf(line, mi->input)) {
+               if (*(mi->content_top) && is_multipart_boundary(mi, line))
+                       return 1;
+       }
+       return 0;
+}
+
+static int handle_boundary(struct mailinfo *mi, struct strbuf *line)
+{
+       struct strbuf newline = STRBUF_INIT;
+
+       strbuf_addch(&newline, '\n');
+again:
+       if (line->len >= (*(mi->content_top))->len + 2 &&
+           !memcmp(line->buf + (*(mi->content_top))->len, "--", 2)) {
+               /* we hit an end boundary */
+               /* pop the current boundary off the stack */
+               strbuf_release(*(mi->content_top));
+               FREE_AND_NULL(*(mi->content_top));
+
+               /* technically won't happen as is_multipart_boundary()
+                  will fail first.  But just in case..
+                */
+               if (--mi->content_top < mi->content) {
+                       error("Detected mismatched boundaries, can't recover");
+                       mi->input_error = -1;
+                       mi->content_top = mi->content;
+                       strbuf_release(&newline);
+                       return 0;
+               }
+               handle_filter(mi, &newline);
+               strbuf_release(&newline);
+               if (mi->input_error)
+                       return 0;
+
+               /* skip to the next boundary */
+               if (!find_boundary(mi, line))
+                       return 0;
+               goto again;
+       }
+
+       /* set some defaults */
+       mi->transfer_encoding = TE_DONTCARE;
+       strbuf_reset(&mi->charset);
+
+       /* slurp in this section's info */
+       while (read_one_header_line(line, mi->input))
+               check_header(mi, line, mi->p_hdr_data, 0);
+
+       strbuf_release(&newline);
+       /* replenish line */
+       if (strbuf_getline_lf(line, mi->input))
+               return 0;
+       strbuf_addch(line, '\n');
+       return 1;
+}
+
+static void handle_body(struct mailinfo *mi, struct strbuf *line)
+{
+       struct strbuf prev = STRBUF_INIT;
+
+       /* Skip up to the first boundary */
+       if (*(mi->content_top)) {
+               if (!find_boundary(mi, line))
+                       goto handle_body_out;
+       }
+
+       do {
+               /* process any boundary lines */
+               if (*(mi->content_top) && is_multipart_boundary(mi, line)) {
+                       /* flush any leftover */
+                       if (prev.len) {
+                               handle_filter(mi, &prev);
+                               strbuf_reset(&prev);
+                       }
+                       if (!handle_boundary(mi, line))
+                               goto handle_body_out;
+               }
+
+               /* Unwrap transfer encoding */
+               decode_transfer_encoding(mi, line);
+
+               switch (mi->transfer_encoding) {
+               case TE_BASE64:
+               case TE_QP:
+               {
+                       struct strbuf **lines, **it, *sb;
+
+                       /* Prepend any previous partial lines */
+                       strbuf_insert(line, 0, prev.buf, prev.len);
+                       strbuf_reset(&prev);
+
+                       /*
+                        * This is a decoded line that may contain
+                        * multiple new lines.  Pass only one chunk
+                        * at a time to handle_filter()
+                        */
+                       lines = strbuf_split(line, '\n');
+                       for (it = lines; (sb = *it); it++) {
+                               if (*(it + 1) == NULL) /* The last line */
+                                       if (sb->buf[sb->len - 1] != '\n') {
+                                               /* Partial line, save it for later. */
+                                               strbuf_addbuf(&prev, sb);
+                                               break;
+                                       }
+                               handle_filter(mi, sb);
+                       }
+                       /*
+                        * The partial chunk is saved in "prev" and will be
+                        * appended by the next iteration of read_line_with_nul().
+                        */
+                       strbuf_list_free(lines);
+                       break;
+               }
+               default:
+                       handle_filter(mi, line);
+               }
+
+               if (mi->input_error)
+                       break;
+       } while (!strbuf_getwholeline(line, mi->input, '\n'));
+
+       flush_inbody_header_accum(mi);
+
+handle_body_out:
+       strbuf_release(&prev);
+}
+
+static void output_header_lines(FILE *fout, const char *hdr, const struct strbuf *data)
+{
+       const char *sp = data->buf;
+       while (1) {
+               char *ep = strchr(sp, '\n');
+               int len;
+               if (!ep)
+                       len = strlen(sp);
+               else
+                       len = ep - sp;
+               fprintf(fout, "%s: %.*s\n", hdr, len, sp);
+               if (!ep)
+                       break;
+               sp = ep + 1;
+       }
+}
+
+static void handle_info(struct mailinfo *mi)
+{
+       struct strbuf *hdr;
+       int i;
+
+       for (i = 0; header[i]; i++) {
+               /* only print inbody headers if we output a patch file */
+               if (mi->patch_lines && mi->s_hdr_data[i])
+                       hdr = mi->s_hdr_data[i];
+               else if (mi->p_hdr_data[i])
+                       hdr = mi->p_hdr_data[i];
+               else
+                       continue;
+
+               if (!strcmp(header[i], "Subject")) {
+                       if (!mi->keep_subject) {
+                               cleanup_subject(mi, hdr);
+                               cleanup_space(hdr);
+                       }
+                       output_header_lines(mi->output, "Subject", hdr);
+               } else if (!strcmp(header[i], "From")) {
+                       cleanup_space(hdr);
+                       handle_from(mi, hdr);
+                       fprintf(mi->output, "Author: %s\n", mi->name.buf);
+                       fprintf(mi->output, "Email: %s\n", mi->email.buf);
+               } else {
+                       cleanup_space(hdr);
+                       fprintf(mi->output, "%s: %s\n", header[i], hdr->buf);
+               }
+       }
+       fprintf(mi->output, "\n");
+}
+
+int mailinfo(struct mailinfo *mi, const char *msg, const char *patch)
+{
+       FILE *cmitmsg;
+       int peek;
+       struct strbuf line = STRBUF_INIT;
+
+       cmitmsg = fopen(msg, "w");
+       if (!cmitmsg) {
+               perror(msg);
+               return -1;
+       }
+       mi->patchfile = fopen(patch, "w");
+       if (!mi->patchfile) {
+               perror(patch);
+               fclose(cmitmsg);
+               return -1;
+       }
+
+       mi->p_hdr_data = xcalloc(MAX_HDR_PARSED, sizeof(*(mi->p_hdr_data)));
+       mi->s_hdr_data = xcalloc(MAX_HDR_PARSED, sizeof(*(mi->s_hdr_data)));
+
+       do {
+               peek = fgetc(mi->input);
+               if (peek == EOF) {
+                       fclose(cmitmsg);
+                       return error("empty patch: '%s'", patch);
+               }
+       } while (isspace(peek));
+       ungetc(peek, mi->input);
+
+       /* process the email header */
+       while (read_one_header_line(&line, mi->input))
+               check_header(mi, &line, mi->p_hdr_data, 1);
+
+       handle_body(mi, &line);
+       fwrite(mi->log_message.buf, 1, mi->log_message.len, cmitmsg);
+       fclose(cmitmsg);
+       fclose(mi->patchfile);
+
+       handle_info(mi);
+       strbuf_release(&line);
+       return mi->input_error;
+}
+
+static int git_mailinfo_config(const char *var, const char *value, void *mi_)
+{
+       struct mailinfo *mi = mi_;
+
+       if (!starts_with(var, "mailinfo."))
+               return git_default_config(var, value, NULL);
+       if (!strcmp(var, "mailinfo.scissors")) {
+               mi->use_scissors = git_config_bool(var, value);
+               return 0;
+       }
+       /* perhaps others here */
+       return 0;
+}
+
+void setup_mailinfo(struct mailinfo *mi)
+{
+       memset(mi, 0, sizeof(*mi));
+       strbuf_init(&mi->name, 0);
+       strbuf_init(&mi->email, 0);
+       strbuf_init(&mi->charset, 0);
+       strbuf_init(&mi->log_message, 0);
+       strbuf_init(&mi->inbody_header_accum, 0);
+       mi->header_stage = 1;
+       mi->use_inbody_headers = 1;
+       mi->content_top = mi->content;
+       git_config(git_mailinfo_config, mi);
+}
+
+void clear_mailinfo(struct mailinfo *mi)
+{
+       int i;
+
+       strbuf_release(&mi->name);
+       strbuf_release(&mi->email);
+       strbuf_release(&mi->charset);
+       strbuf_release(&mi->inbody_header_accum);
+       free(mi->message_id);
+
+       if (mi->p_hdr_data)
+               for (i = 0; mi->p_hdr_data[i]; i++)
+                       strbuf_release(mi->p_hdr_data[i]);
+       free(mi->p_hdr_data);
+       if (mi->s_hdr_data)
+               for (i = 0; mi->s_hdr_data[i]; i++)
+                       strbuf_release(mi->s_hdr_data[i]);
+       free(mi->s_hdr_data);
+
+       while (mi->content < mi->content_top) {
+               free(*(mi->content_top));
+               mi->content_top--;
+       }
+
+       strbuf_release(&mi->log_message);
+}
diff --git a/t/t4256/1/patch b/t/t4256/1/patch
new file mode 100644 (file)
index 0000000..bd0d8b0
--- /dev/null
@@ -0,0 +1,129 @@
+From: A <author@example.com>
+Subject: [PATCH] mailinfo: support format=flowed
+Message-ID: <aaaaaaaa-aaaa-aaaa-aaaa-aaaaaaaaaaaa@example.com>
+Date: Sat, 25 Aug 2018 22:04:50 +0200
+User-Agent: Mozilla/5.0 (Windows NT 10.0; WOW64; rv:60.0) Gecko/20100101
+ Thunderbird/60.0
+MIME-Version: 1.0
+Content-Type: text/plain; charset=utf-8; format=flowed
+Content-Language: en-US
+Content-Transfer-Encoding: 7bit
+
+---
+  mailinfo.c | 64 ++++++++++++++++++++++++++++++++++++++++++++++++++++--
+  1 file changed, 62 insertions(+), 2 deletions(-)
+
+diff --git a/mailinfo.c b/mailinfo.c
+index 3281a37d51..b395adbdf2 100644
+--- a/mailinfo.c
++++ b/mailinfo.c
+@@ -237,11 +237,22 @@ static int slurp_attr(const char *line, const char 
+*name, struct strbuf *attr)
+       return 1;
+  }
+
++static int has_attr_value(const char *line, const char *name, const 
+char *value)
++{
++      struct strbuf sb = STRBUF_INIT;
++      int rc = slurp_attr(line, name, &sb) && !strcasecmp(sb.buf, value);
++      strbuf_release(&sb);
++      return rc;
++}
++
+  static void handle_content_type(struct mailinfo *mi, struct strbuf *line)
+  {
+       struct strbuf *boundary = xmalloc(sizeof(struct strbuf));
+       strbuf_init(boundary, line->len);
+
++      mi->format_flowed = has_attr_value(line->buf, "format=", "flowed");
++      mi->delsp = has_attr_value(line->buf, "delsp=", "yes");
++
+       if (slurp_attr(line->buf, "boundary=", boundary)) {
+               strbuf_insert(boundary, 0, "--", 2);
+               if (++mi->content_top >= &mi->content[MAX_BOUNDARIES]) {
+@@ -964,6 +975,52 @@ static int handle_boundary(struct mailinfo *mi, 
+struct strbuf *line)
+       return 1;
+  }
+
++static void handle_filter_flowed(struct mailinfo *mi, struct strbuf *line,
++                               struct strbuf *prev)
++{
++      size_t len = line->len;
++      const char *rest;
++
++      if (!mi->format_flowed) {
++              handle_filter(mi, line);
++              return;
++      }
++
++      if (line->buf[len - 1] == '\n') {
++              len--;
++              if (len && line->buf[len - 1] == '\r')
++                      len--;
++      }
++
++      /* Keep signature separator as-is. */
++      if (skip_prefix(line->buf, "-- ", &rest) && rest - line->buf == len) {
++              if (prev->len) {
++                      handle_filter(mi, prev);
++                      strbuf_reset(prev);
++              }
++              handle_filter(mi, line);
++              return;
++      }
++
++      /* Unstuff space-stuffed line. */
++      if (len && line->buf[0] == ' ') {
++              strbuf_remove(line, 0, 1);
++              len--;
++      }
++
++      /* Save flowed line for later, but without the soft line break. */
++      if (len && line->buf[len - 1] == ' ') {
++              strbuf_add(prev, line->buf, len - !!mi->delsp);
++              return;
++      }
++
++      /* Prepend any previous partial lines */
++      strbuf_insert(line, 0, prev->buf, prev->len);
++      strbuf_reset(prev);
++
++      handle_filter(mi, line);
++}
++
+  static void handle_body(struct mailinfo *mi, struct strbuf *line)
+  {
+       struct strbuf prev = STRBUF_INIT;
+@@ -1012,7 +1069,7 @@ static void handle_body(struct mailinfo *mi, 
+struct strbuf *line)
+                                               strbuf_addbuf(&prev, sb);
+                                               break;
+                                       }
+-                              handle_filter(mi, sb);
++                              handle_filter_flowed(mi, sb, &prev);
+                       }
+                       /*
+                        * The partial chunk is saved in "prev" and will be
+@@ -1022,13 +1079,16 @@ static void handle_body(struct mailinfo *mi, 
+struct strbuf *line)
+                       break;
+               }
+               default:
+-                      handle_filter(mi, line);
++                      handle_filter_flowed(mi, line, &prev);
+               }
+
+               if (mi->input_error)
+                       break;
+       } while (!strbuf_getwholeline(line, mi->input, '\n'));
+
++      if (prev.len)
++              handle_filter(mi, &prev);
++
+       flush_inbody_header_accum(mi);
+
+  handle_body_out:
+-- 
+2.18.0