Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Tue, 20 Dec 2016 15:13:39 -0500
From: Rich Felker <dalias@...c.org>
To: musl@...ts.openwall.com
Subject: Re: [PATCH] implement strftime GNU extension padding
 specifiers '_', '-' and '0'

On Tue, Nov 22, 2016 at 10:29:08AM +0200, Timo Teräs wrote:
> ---
> For review. GNU and most BSDs seem to implement these extensions.

>From what I can tell, only FreeBSD does and they call it a GNU
extension. OpenBSD does not have them at all, and also lacks the
POSIX-mandated width specifiers. Not sure if FreeBSD has those.

> This applies them to numbers only. There might be few cases where this
> could be applied more, but I think covers most of the uses. Is there any
> test suite I could verify this against?
> 
>  src/time/strftime.c | 22 ++++++++++++++--------
>  1 file changed, 14 insertions(+), 8 deletions(-)

I'm not aware of any test suite. Test cases for strftime (standard
features especially -- to make sure none of them are broken by
changes) would be really nice to have in libc-test.

Overall I'm not too excited about adding nonstandard extensions like
this unless there's indication that there's some good chance of
consensus among implementations. They have the potential to conflict
with future standard requirements and there's no reliable way to test
for them at build time (unlike nonstd functions where you can at least
check if the symbol is present). So the main thing I'd like before we
move forward with this is research into whether other implementations
support these already, and if not, what their attitudes towards it
are. OTOH if the BSDs aren't even going to adopt the POSIX
requirements then maybe they're not terribly relevant here.

As we discussed on IRC, python documents that these are nonstandard
and depend on your system's strftime, so adding them just for python
does not seem justified.

I'm not saying no to this outright but I think it should have some
more thought before deciding one way or the other.

Rich

> diff --git a/src/time/strftime.c b/src/time/strftime.c
> index e103e02..14fe6f5 100644
> --- a/src/time/strftime.c
> +++ b/src/time/strftime.c
> @@ -48,12 +48,12 @@ static int week_num(const struct tm *tm)
>  const char *__tm_to_tzname(const struct tm *);
>  size_t __strftime_l(char *restrict, size_t, const char *restrict, const struct tm *restrict, locale_t);
>  
> -const char *__strftime_fmt_1(char (*s)[100], size_t *l, int f, const struct tm *tm, locale_t loc)
> +const char *__strftime_fmt_1(char (*s)[100], size_t *l, int f, const struct tm *tm, locale_t loc, int pad)
>  {
>  	nl_item item;
>  	long long val;
>  	const char *fmt = "-";
> -	int width = 2;
> +	int width = 2, def_pad = '0';
>  
>  	switch (f) {
>  	case 'a':
> @@ -79,15 +79,14 @@ const char *__strftime_fmt_1(char (*s)[100], size_t *l, int f, const struct tm *
>  	case 'C':
>  		val = (1900LL+tm->tm_year) / 100;
>  		goto number;
> +	case 'e':
> +		def_pad = '_';
>  	case 'd':
>  		val = tm->tm_mday;
>  		goto number;
>  	case 'D':
>  		fmt = "%m/%d/%y";
>  		goto recu_strftime;
> -	case 'e':
> -		*l = snprintf(*s, sizeof *s, "%2d", tm->tm_mday);
> -		return *s;
>  	case 'F':
>  		fmt = "%Y-%m-%d";
>  		goto recu_strftime;
> @@ -200,7 +199,12 @@ const char *__strftime_fmt_1(char (*s)[100], size_t *l, int f, const struct tm *
>  		return 0;
>  	}
>  number:
> -	*l = snprintf(*s, sizeof *s, "%0*lld", width, val);
> +	switch (pad ? pad : def_pad) {
> +	case '-': *l = snprintf(*s, sizeof *s, "%lld", val); break;
> +	case '_': *l = snprintf(*s, sizeof *s, "%*lld", width, val); break;
> +	case '0':
> +	default:  *l = snprintf(*s, sizeof *s, "%0*lld", width, val); break;
> +	}
>  	return *s;
>  nl_strcat:
>  	fmt = __nl_langinfo_l(item, loc);
> @@ -221,7 +225,7 @@ size_t __strftime_l(char *restrict s, size_t n, const char *restrict f, const st
>  	char buf[100];
>  	char *p;
>  	const char *t;
> -	int plus;
> +	int pad, plus;
>  	unsigned long width;
>  	for (l=0; l<n; f++) {
>  		if (!*f) {
> @@ -233,6 +237,8 @@ size_t __strftime_l(char *restrict s, size_t n, const char *restrict f, const st
>  			continue;
>  		}
>  		f++;
> +		pad = 0;
> +		if (*f == '-' || *f == '_' || *f == '0') pad = *f++;
>  		if ((plus = (*f == '+'))) f++;
>  		width = strtoul(f, &p, 10);
>  		if (*p == 'C' || *p == 'F' || *p == 'G' || *p == 'Y') {
> @@ -242,7 +248,7 @@ size_t __strftime_l(char *restrict s, size_t n, const char *restrict f, const st
>  		}
>  		f = p;
>  		if (*f == 'E' || *f == 'O') f++;
> -		t = __strftime_fmt_1(&buf, &k, *f, tm, loc);
> +		t = __strftime_fmt_1(&buf, &k, *f, tm, loc, pad);
>  		if (!t) break;
>  		if (width) {
>  			for (; *t=='+' || *t=='-' || (*t=='0'&&t[1]); t++, k--);
> -- 
> 2.10.2

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.