printf("%m") doesn't generate a warning -- shouldn't it?

[Dimitry Andric]

On 21 May 2018, at 15:43, Konstantin Belousov <kostikbel at gmail.com> wrote:
> 
> On Mon, May 21, 2018 at 01:19:01PM +0200, Dimitry Andric wrote:
>> On 21 May 2018, at 04:38, Thomas Munro <munro at ip9.org> wrote:
>>> 
>>> As discussed on the PostgreSQL[1] and NetBSD mailing lists[2][3],
>>> syslog-like printf("%m") is a GNU extension that doesn't generate a
>>> warning from Clang or GCC on other operating systems even though when
>>> it doesn't actually work.  That's because the __printf__ attribute
>>> that our __printflike macro in /usr/include/stdio.h expands to
>>> effectively means "like printf in glibc, allowing %m", not like POSIX
>>> or our actual libc which just prints out "m" when it sees it.
>>> 
>>> It'd sure be nice to get a compiler warning on FreeBSD when porting
>>> software that uses that if it doesn't actually work (or ... to support
>>> it).
>> 
>> Please submit upstream bug(s) for clang and gcc.  It turns out clang has
>> a -Wformat-non-iso warning flag which should warn about this, but I
>> can't get it to emit it:
>> 
>> $ cat printf-m.c
>> #include <stdio.h>
>> 
>> int main(void)
>> {
>>        printf("error: %m\n");
>>        return 0;
>> }
>> 
>> $ clang -std=c99 -Wformat-non-iso -c printf-m.c
>> <nothing>
> 
> Why not add %m instead ?  It is very easy and several people did it in
> round-about ways.

Sure, that is certainly a nice thing to have, and let's commit it.  But
it's still an upstream bug if an explicit warning flag for non-ISO
printf specifiers doesn't work.  I'll take that discussion to the LLVM
bug tracker.


> diff --git a/lib/libc/stdio/vfprintf.c b/lib/libc/stdio/vfprintf.c
> index 29fbd95b399..70f5074e2f7 100644
> --- a/lib/libc/stdio/vfprintf.c
> +++ b/lib/libc/stdio/vfprintf.c
> @@ -317,6 +317,7 @@ __vfprintf(FILE *fp, locale_t locale, const char *fmt0, va_list ap)
> 	int ret;		/* return value accumulator */
> 	int width;		/* width from format (%8d), or 0 */
> 	int prec;		/* precision from format; <0 for N/A */
> +	int saved_errno;
> 	char sign;		/* sign prefix (' ', '+', '-', or \0) */
> 	struct grouping_state gs; /* thousands' grouping info */
> 
> @@ -466,6 +467,7 @@ __vfprintf(FILE *fp, locale_t locale, const char *fmt0, va_list ap)
> 	savserr = fp->_flags & __SERR;
> 	fp->_flags &= ~__SERR;
> 
> +	saved_errno = errno;
> 	convbuf = NULL;
> 	fmt = (char *)fmt0;
> 	argtable = NULL;
> @@ -776,6 +778,11 @@ reswitch:	switch (ch) {
> 			}
> 			break;
> #endif /* !NO_FLOATING_POINT */
> +		case 'm':
> +			cp = strerror(saved_errno);
> +			size = (prec >= 0) ? strnlen(cp, prec) : strlen(cp);
> +			sign = '\0';
> +			break;
> 		case 'n':
> 			/*
> 			 * Assignment-like behavior is specified if the

LGTM.

-Dimitry

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 223 bytes
Desc: Message signed with OpenPGP
URL: <http://lists.freebsd.org/pipermail/freebsd-hackers/attachments/20180522/e98f60e6/attachment.sig>