From: Junio C Hamano <junkio@cox.net>
Date: Tue, 20 Feb 2007 09:04:25 GMT
Subject: Re: [PATCH] Change "refs/" references to symbolic constants
Message-ID: <7vlkitxkrq.fsf@assigned-by-dhcp.cox.net>
In-Reply-To: <200702200841.26578.andyparkins@gmail.com>

Andy Parkins <andyparkins@gmail.com> writes:

> As I've said before, I'm against /any/ special treatment of
> master other than as a default branch name in a newly
> initialised repository.  PATH_REFS_HEADS_MASTER is closer to
> master not being special than without so I'd be in favour of
> that.

Ok.

>> These repeated strncmp(p, X, STRLEN_X) almost makes me wonder if we
>> want to introduce:
>>
>> 	inline int prefixcmp(a, b)
>>         {
>>         	return (strncmp(a, b, strlen(b));
>>         }
>>
>> with clever preprocessor optimization to have compiler do strlen()
>> when b is a string literal.
>
> Wow; that would be clever - regardless of whether this patch
> is acceptable or not.

Actually GCC seems to be clever enough.  It appears that I do
not even have to do prefixcmp_0() below; prefixcmp_1() seems to
generate good code, with GCC 4.1.2 prerelease on my x86_64.

-- >8 --
#include <string.h>
static inline int prefixcmp_0(const char *a, const char *b)
{
	if (__builtin_constant_p(b))
		return strncmp(a, b, sizeof(b) - 1);
	else
		return strncmp(a, b, strlen(b));
}

static inline prefixcmp_1(const char *a, const char *b)
{
	return strncmp(a, b, strlen(b));
}

void foo(const char *s, const char *t, int *a, int *b, int *c, int *d)
{
	*a = prefixcmp_0(s, "abcdefg");
	*b = prefixcmp_0(s, t);
	*c = prefixcmp_1(s, "ABCDEFGH");
	*d = prefixcmp_1(s, t);
}
-- 8< --