Re*: Content-Type handling

[Junio C Hamano <gitster@pobox.com>]

Tomas Cohen Arazi <tomascohen@gmail.com> writes:

> Hi, I'm not sure it is a bug, but we used:
>
> git config --global format.headers "Content-Type: text/plain; charset=\"utf-8\""
>
> and recently (perhaps an Ubuntu default setup issue) the content-type
> is being automatically set, the result is that patches contain this:
>
> Content-Type: text/plain; charset=UTF-8
> Content-Transfer-Encoding: 8bit
> Content-Type: text/plain; charset="utf-8"
>
> This might not be the problem, but when I apply the patch I get this:
>
> fatal: cannot convert from UTF-8utf-8 to UTF-8
>
> which looks like a bug. Not sure it hasn't been reported before, but I
> think it should take one of the content-type specifications and not
> append both. I couldn't find a place to looks for previous bug reports
> to check, so forgive me if this is not the right place to report it.

What I think is happening is:

 * ancient "git format-patch" did not add appropriate content-type
   headers, and that bug was fixed in more recent versions.

 * while waiting for the bug to get fixed, the user worked around by
   having her own format.headers to work it around.

 * the user upgraded to a newer version of git without the bug.

 * Git trusts that the user knows better than git itself, and it
   adds format.headers to whatever it needs to have in the message.
   As a message with more than one content-type: is invalid, the
   handcrafted format.headers used to be used as a workaround should
   be removed, but it hasn't been.

 * Then the receiving end (mailinfo) gets confused by the invalid
   input.


At least, a patch line this may alleviate the symptom.  It seems
that we assumed we won't see multiple charset="..." so the parser,
starting from an empty charset, simply appended whatever it found
in the message, instead of overwriting the old value it had.  The
other user of the function this patch touches always makes sure the
"attr" strbuf is empty before calling the function, and wouldn't
have been confused by multiple occurrences of the same header, and
this patch should not affect it, either.

 builtin/mailinfo.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git i/builtin/mailinfo.c w/builtin/mailinfo.c
index b691b77..2b3f4d9 100644
--- i/builtin/mailinfo.c
+++ w/builtin/mailinfo.c
@@ -160,10 +160,9 @@ static int slurp_attr(const char *line, const char *name, struct strbuf *attr)
 	const char *ends, *ap = strcasestr(line, name);
 	size_t sz;
 
-	if (!ap) {
-		strbuf_setlen(attr, 0);
+	strbuf_setlen(attr, 0);
+	if (!ap)
 		return 0;
-	}
 	ap += strlen(name);
 	if (*ap == '"') {
 		ap++;