From: Junio C Hamano <gitster@pobox.com>
Date: Tue, 13 Jan 2009 08:13:18 GMT
Subject: Re: [PATCH 2/2] grep: don't call regexec() for fixed strings
Message-ID: <7vy6xfd3oh.fsf@gitster.siamese.dyndns.org>
In-Reply-To: <496B9780.3030000@lsrfire.ath.cx>

René Scharfe <rene.scharfe@lsrfire.ath.cx> writes:

> +#define RR GIT_REGEX_SPECIAL /* $, (, ), +, ., ^, {, | */
> +#define US GIT_UNDERSCORE
> +#define Ah (GIT_ALPHA | GIT_LOWER_XDIGIT)
>  
>  unsigned char sane_ctype[256] = {
>  	GS,  0,  0,  0,  0,  0,  0,  0,  0, SS, SS,  0,  0, SS,  0,  0,		/* 0-15 */

Mental note.  NUL is marked as GIT_SPECIAL.

>  #define isspecial(x) sane_istest(x,GIT_SPECIAL)
> +#define isregexspecial(x) sane_istest(x,GIT_SPECIAL | GIT_REGEX_SPECIAL)

Perhaps isspecial() is misnamed if we were to enhance git-ctype in this
way.  It is about a byte being shell glob pattern or a NUL (!!!), and it
should be renamed to isglobspecial() or something.

dir.c uses isspecial() in two places, and both callers rely on NUL being a
part of special to terminate the loops they are in, like this:

	for (;;) {
		unsigned char c = *match++;
		len++;
		if (isspecial(c))
			return len;
	}

It may be a cunning and cute logic, but I do not particularly like it.  It
might be cleaner to rename it to isglobspecial(), drop NUL from it, and
have these two call existing call sites to explicitly check for (c == NUL)
for loop termination.