Problem in Patches with commit-messages containing non-ascii

Hello

I stumbled over a problem with git handling patches.
Perhaps i am then only developer who hast this problem because im am the only developer writing commit-mesages in german.

Consider this log-Message:
commit ea2cd63dfe9b3ac3581b6cff8b13a52e69066242

Author: martin <[email protected]>
Date:   Fri Nov 19 18:58:58 2010 +0100

Methoden überall angepasst.
Ausser Aussnahmen

Using format-patch the result is:

From ea2cd63dfe9b3ac3581b6cff8b13a52e69066242 Mon Sep 17 00:00:00 2001
From: martin <[email protected]>
Date: Fri, 19 Nov 2010 18:58:58 +0100
Subject: [PATCH] =?UTF-8?q?Methoden=20=C3=BCberall=20angepasst. =20Ausser=20Aussnahmen?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

The content of the subject field ist split up in two lines . The blank in the second line indicates an header-folding according to RFC2822 .
After this the string is encoded according to RFC2047 because it contains non-ascii chars. The blank indicatinc the folding ist encoded too, with =20.
That is a Problem because the unfolding according to RFC2822 cant't detect the folding anymore. RFC2822 suggests that the unfolding must be done before any further processing of the header which applies to the RFC2047 decoding.

Applying this patch leads to this commit-Message:

commit 3949e57e4773e85e6c55482b68ade7c409426b3c

Author: martin <[email protected]>
Date:   Fri Nov 19 18:58:58 2010 +0100

    =?UTF-8?q?Methoden=20=C3=BCberall=20angepasst.

    =20Ausser=20Aussnahmen?=
    MIME-Version: 1.0
    Content-Type: text/plain; charset=UTF-8
    Content-Transfer-Encoding: 8bit

The solution is to make an exception not to encode blanks indicating a folding.

I wrote this patch:

index f85444b..8a78a4e 100644
@@ -216,7 +216,7 @@ static int is_rfc2047_special(char ch)
 static void add_rfc2047(struct strbuf *sb, const char *line, int len,
 		       const char *encoding)
 {
-	int i, last;
+	int i, last, num_foldings;
 
 	for (i = 0; i < len; i++) {
 		int ch = line[i];

@@ -229,8 +229,14 @@ static void add_rfc2047(struct strbuf *sb, const

char *line, int len,
return;

needquote:
- strbuf_grow(sb, len * 3 + strlen(encoding) + 100);
+ num_foldings=0;
+ for (i = 1; i < len; i++)
+ if(line[i]==' '&&line[i]=='\n')
+ num_foldings++;
+
+ strbuf_grow(sb, len * 3 + num_foldings*(7+strlen(encoding)) + 100);
strbuf_addf(sb, "=?%s?q?", encoding);
+ unsigned last_ch=0;
for (i = last = 0; i < len; i++) {
unsigned ch = line[i] & 0xFF;
/*
@@ -240,10 +246,19 @@ needquote:
* leave the underscore in place. */
if (is_rfc2047_special(ch) || ch == ' ') {
- strbuf_add(sb, line + last, i - last);
- strbuf_addf(sb, "=%02X", ch);
- last = i + 1;
+ if(!(ch == ' '&& last_ch=='\n')){
+ strbuf_add(sb, line + last, i - last);
+ strbuf_addf(sb, "=%02X", ch);
+ }
+ else{
+ if(i>last+1)
+ strbuf_add(sb, line + last, i - last-1);
+ strbuf_addstr(sb, "?=\n ");
+ strbuf_addf(sb, "=?%s?q?", encoding);
+ }
+ last = i + 1;
}
+ last_ch=ch;
}
strbuf_add(sb, line + last, len - last); strbuf_addstr(sb, "?=");

Then git generates this patch:

From ea2cd63dfe9b3ac3581b6cff8b13a52e69066242 Mon Sep 17 00:00:00 2001
From: martin <[email protected]>
Date: Fri, 19 Nov 2010 18:58:58 +0100
Subject: [PATCH] =?UTF-8?q?Methoden=20=C3=BCberall=20angepasst.?=
=?UTF-8?q?Ausser=20Aussnahmen?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Applyin leads to a correct commit-Messsage:

commit 62d06e3415ec0726dbd58c11ed93771502b77805

Author: martin <[email protected]>
Date:   Fri Nov 19 18:58:58 2010 +0100

Methoden überall angepasst.Ausser Aussnahmen

Best regards
martin

FWIW, support for multi-line summaries is very limited. Several tools assume that the log message has this format:

<Summary in one line>
<Blank line>
<Details>

So one could argue that your patch fixes something that isn't really supported anyway.

How is that correct? It's different from the original commit message.

-Jan

Jan Krüger venit, vidit, dixit 03.12.2010 13:59:

Also, it is "Außer Ausnahmen" even after the latest spelling reform ;)

Michael

That's at least missing a space after the period.

Andreas.

Andreas Schwab, [email protected]
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."