Re: [PATCH 3/3] implement pattern matching in ce_path_match

From: Junio C Hamano <gitster@pobox.com>
Date: Jan 14, 2009, 18:39 UTC
Message-ID: <7vljtd20m6.fsf@gitster.siamese.dyndns.org>
In-Reply-To: <1231944876-29930-4-git-send-email-drizzd@aon.at>

Clemens Buchacher <drizzd@aon.at> writes:

> With this patch ce_path_match uses match_pathspec in order to perform
> pattern matching.

We have two conflicting definitions of pattern matching in our system. I'd make it more explicit which kind of pattern matching you are talking about here.

The family of operations based on the diff-tree machinery (e.g. path limited revision walking "git log A..B -- dir1/dir2") define the pattern matching as "leading path match (exact match is just a special case of this)". Other operations that work on paths in the work tree and the index (e.g. grep, ls-files) uses "leading path match, but fall back to globbing".

In the longer term we really should unify them by teaching the former to fall back to globbing without getting undue performance hit, and this patch may be a step in the right direction. There are optimizations that assume the "leading path" semantics to trim the input early and avoid opening and descending into a tree object if pathspec patterns cannot possibly match (see tree-diff.c::tree_entry_interesting() for an example), and we need to teach them to notice a glob wildcard in an earlier part of a pathspec and to descend into some trees that they would have skipped with the old definition of pathspec.

Show 8 quoted lines

> @@ -49,7 +60,7 @@ static void cmd_log_init(int argc, const char **argv, const char *prefix,
>  		rev->always_show_header = 0;
>  	if (DIFF_OPT_TST(&rev->diffopt, FOLLOW_RENAMES)) {
>  		rev->always_show_header = 0;
> -		if (rev->diffopt.nr_paths != 1)
> +		if (rev->diffopt.nr_paths != 1 || has_special(rev->diffopt.paths[0]))
>  			usage("git logs can only follow renames on one pathname at a time");
>  	}

The reason match_pathspec() first tries exact match and then falls back to globbing is so that the user can say "I have a file whose name ends with a question mark, please match it literally." This patch defeats it, but it probably is a minor point.

1/3 and 2/3 in the series looked good.

Thanks.