Skip to content

Commit 9371c0e

Browse files
avargitster
authored andcommitted
gettext.c: remove/reword a mostly-useless comment
Mostly remove the comment I added 5e9637c (i18n: add infrastructure for translating Git with gettext, 2011-11-18). Since then we had a fix in 9c0495d (gettext.c: detect the vsnprintf bug at runtime, 2013-12-01) so we're not running with the "set back to C locale" hack on any modern system. So having more than 1/4 of the file taken up by a digression about a glibc bug that mostly doesn't happen to anyone anymore is just a needless distraction. Shorten the comment to make a brief mention of the bug, and where to find more info by looking at the git history for this now-removed comment. Signed-off-by: Ævar Arnfjörð Bjarmason <[email protected]> Signed-off-by: Junio C Hamano <[email protected]>
1 parent 450d740 commit 9371c0e

File tree

1 file changed

+15
-79
lines changed

1 file changed

+15
-79
lines changed

gettext.c

Lines changed: 15 additions & 79 deletions
Original file line numberDiff line numberDiff line change
@@ -87,88 +87,24 @@ static int test_vsnprintf(const char *fmt, ...)
8787

8888
static void init_gettext_charset(const char *domain)
8989
{
90-
/*
91-
This trick arranges for messages to be emitted in the user's
92-
requested encoding, but avoids setting LC_CTYPE from the
93-
environment for the whole program.
94-
95-
This primarily done to avoid a bug in vsnprintf in the GNU C
96-
Library [1]. which triggered a "your vsnprintf is broken" error
97-
on Git's own repository when inspecting v0.99.6~1 under a UTF-8
98-
locale.
99-
100-
That commit contains a ISO-8859-1 encoded author name, which
101-
the locale aware vsnprintf(3) won't interpolate in the format
102-
argument, due to mismatch between the data encoding and the
103-
locale.
104-
105-
Even if it wasn't for that bug we wouldn't want to use LC_CTYPE at
106-
this point, because it'd require auditing all the code that uses C
107-
functions whose semantics are modified by LC_CTYPE.
108-
109-
But only setting LC_MESSAGES as we do creates a problem, since
110-
we declare the encoding of our PO files[2] the gettext
111-
implementation will try to recode it to the user's locale, but
112-
without LC_CTYPE it'll emit something like this on 'git init'
113-
under the Icelandic locale:
114-
115-
Bj? til t?ma Git lind ? /hlagh/.git/
116-
117-
Gettext knows about the encoding of our PO file, but we haven't
118-
told it about the user's encoding, so all the non-US-ASCII
119-
characters get encoded to question marks.
120-
121-
But we're in luck! We can set LC_CTYPE from the environment
122-
only while we call nl_langinfo and
123-
bind_textdomain_codeset. That suffices to tell gettext what
124-
encoding it should emit in, so it'll now say:
125-
126-
Bjó til tóma Git lind í /hlagh/.git/
127-
128-
And the equivalent ISO-8859-1 string will be emitted under a
129-
ISO-8859-1 locale.
130-
131-
With this change way we get the advantages of setting LC_CTYPE
132-
(talk to the user in his language/encoding), without the major
133-
drawbacks (changed semantics for C functions we rely on).
134-
135-
However foreign functions using other message catalogs that
136-
aren't using our neat trick will still have a problem, e.g. if
137-
we have to call perror(3):
138-
139-
#include <stdio.h>
140-
#include <locale.h>
141-
#include <errno.h>
142-
143-
int main(void)
144-
{
145-
setlocale(LC_MESSAGES, "");
146-
setlocale(LC_CTYPE, "C");
147-
errno = ENODEV;
148-
perror("test");
149-
return 0;
150-
}
151-
152-
Running that will give you a message with question marks:
153-
154-
$ LANGUAGE= LANG=de_DE.utf8 ./test
155-
test: Kein passendes Ger?t gefunden
156-
157-
The vsnprintf bug has been fixed since glibc 2.17.
158-
159-
Then we could simply set LC_CTYPE from the environment, which would
160-
make things like the external perror(3) messages work.
161-
162-
See t/t0203-gettext-setlocale-sanity.sh's "gettext.c" tests for
163-
regression tests.
164-
165-
1. http://sourceware.org/bugzilla/show_bug.cgi?id=6530
166-
2. E.g. "Content-Type: text/plain; charset=UTF-8\n" in po/is.po
167-
*/
16890
setlocale(LC_CTYPE, "");
16991
charset = locale_charset();
17092
bind_textdomain_codeset(domain, charset);
171-
/* the string is taken from v0.99.6~1 */
93+
94+
/*
95+
* Work around an old bug fixed in glibc 2.17 (released on
96+
* 2012-12-24), at the cost of potentially making translated
97+
* messages from external functions like perror() emitted in
98+
* the wrong encoding.
99+
*
100+
* The bug affected e.g. git.git's own 7eb93c89651 ([PATCH]
101+
* Simplify git script, 2005-09-07), which is the origin of
102+
* the "David_K\345gedal" test string.
103+
*
104+
* See a much longer comment added to this file in 5e9637c6297
105+
* (i18n: add infrastructure for translating Git with gettext,
106+
* 2011-11-18) for more details.
107+
*/
172108
if (test_vsnprintf("%.*s", 13, "David_K\345gedal") < 0)
173109
setlocale(LC_CTYPE, "C");
174110
}

0 commit comments

Comments
 (0)