git/list[1] front-page[2] threads[3] people[4] search[5] about
 

[PATCH 4/4] exclude: reuse last basename comparison

From
Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Date
Jun 7, 2012, 07:53 UTC
Message-ID
<1339055617-23028-4-git-send-email-pclouds@gmail.com>
In-Reply-To
<1339055617-23028-1-git-send-email-pclouds@gmail.com>

If two consecutive patterns share the same "base" and we just compared base against pathname last time, we can just reuse the last comparison result.

This optmization is made with read_directory() in mind. Notice that all exclude patterns share the same "base" pointer, which is basebuf[] from "struct dir" (given indirectly by prep_exclude()) and patterns from the same .gitignore will stay in the same order. This opens an opportunity for this optimization when there are a lot of patterns in subdirectories-with-long-path-name/.gitignore.

Other users of excluded_from_list() unlikely take advantage of this, unless add_excludes() learns to pre-compare two consecutive bases and save the result, so excluded_from_list() can perform a cheap "are these two bases the same" check.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
---
 I haven't put more thought/work on optimizing the big top .gitignore
 case yet.  But something like this is probably worth doing anyway.
 Pathspec might use the same optimization. If a user does (notice no
 quotes)
 
 git something long/path/to/here/*.[ch]
 we would need to compare "long/path/to/here" all over again (I
 haven't checked the code though).
 dir.c | 23 ++++++++++++++++++++---
 1 file changed, 20 insertions(+), 3 deletions(-)
diff --git a/dir.c b/dir.c
index 94fe9f8..2964076 100644
--- a/dir.c
+++ b/dir.c
@@ -511,6 +511,7 @@ int excluded_from_list(const char *pathname,
 		       struct exclude_list *el)
 {
 	int i;
+	int last_basecmp = el->nr, basecmp_result;
 
 	if (!el->nr)
 		return -1;	/* undefined */
@@ -552,9 +553,25 @@ int excluded_from_list(const char *pathname,
 			prefix--;
 		}
 
-		if (pathlen < x->baselen ||
-		    (x->baselen && pathname[x->baselen-1] != '/') ||
-		    strncmp_icase(pathname, x->base, x->baselen))
+		if (i < el->nr - 1 &&
+		    last_basecmp == i + 1 &&
+		    x->base    == el->excludes[last_basecmp]->base &&
+		    x->baselen == el->excludes[last_basecmp]->baselen)
+			/*
+			 * we have the same "base" as last time and
+			 * last time we came here too (i.e. no break
+			 * or continue from the above code), reuse
+			 * basecmp_result
+			 */
+			;
+		else
+			/* anything other than zero is ok, we don't
+			   really care about the sorting order */
+			basecmp_result = pathlen < x->baselen ||
+				(x->baselen && pathname[x->baselen - 1] != '/') ||
+				strncmp_icase(pathname, x->base, x->baselen);
+		last_basecmp = i;
+		if (basecmp_result)
 			continue;
 
 		namelen = x->baselen ? pathlen - x->baselen : pathlen;
-- 
1.7.11.rc1.185.g281ad67
Previous: Nguyễn Thái Ngọc DuyNext: Junio C Hamano
Message 4 of 7 in “Unindent excluded_from_list()”
  1. 1/4 Unindent excluded_from_list()Nguyễn Thái Ngọc Duy, Jun 7, 2012
  2. 2/4 dir.c: get rid of the wildcard symbol set in no_wildcard()Nguyễn Thái Ngọc Duy, Jun 7, 2012
  3. 3/4 exclude: do strcmp as much as possible before fnmatchNguyễn Thái Ngọc Duy, Jun 7, 2012
  4. 4/4 exclude: reuse last basename comparisonNguyễn Thái Ngọc Duy, Jun 7, 2012
  5. Junio C HamanoJun 7, 2012
  6. Junio C HamanoJun 7, 2012
  7. Nguyen Thai Ngoc DuyJun 8, 2012

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.