git.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Stephan Beyer <s-beyer@gmx.net>
To: git@vger.kernel.org
Cc: Stephan Beyer <s-beyer@gmx.net>,
	Christian Couder <chriscool@tuxfamily.org>
Subject: [PATCH 05/16] bisect: get rid of recursion in count_distance()
Date: Fri, 26 Feb 2016 03:04:31 +0100	[thread overview]
Message-ID: <1456452282-10325-6-git-send-email-s-beyer@gmx.net> (raw)
In-Reply-To: <1456452282-10325-1-git-send-email-s-beyer@gmx.net>

Large repositories with a huge amount of merge commits in the
bisection process could lead to stack overflows in git bisect.
In order to prevent this, this commit uses an *iterative* version
for counting the number of ancestors of a commit.

Signed-off-by: Stephan Beyer <s-beyer@gmx.net>
---

In the cover letter I promised that bisect will be faster.
It's not rocket science to know that replacing a simple recursion
by an iterative version involving a list (with alloc'ing and stuff)
makes the code slower.

For my git master branch test, this changed the running time
from ~50 to ~60 seconds. However, for a smaller stack size,
this changes the running time from <crash> to ~60 seconds.
That's a win ;-)

 bisect.c | 55 ++++++++++++++++++++++---------------------------------
 1 file changed, 22 insertions(+), 33 deletions(-)

diff --git a/bisect.c b/bisect.c
index 412e2c0..03e7660 100644
--- a/bisect.c
+++ b/bisect.c
@@ -26,33 +26,23 @@ static const char *term_good;
 /* Remember to update object flag allocation in object.h */
 #define COUNTED		(1u<<16)
 
-/*
- * This is a truly stupid algorithm, but it's only
- * used for bisection, and we just don't care enough.
- *
- * We care just barely enough to avoid recursing for
- * non-merge entries.
- */
 static int count_distance(struct commit_list *entry)
 {
 	int nr = 0;
+	struct commit_list *todo = NULL;
+	commit_list_append(entry->item, &todo);
 
-	while (entry) {
-		struct commit *commit = entry->item;
-		struct commit_list *p;
+	while (todo) {
+		struct commit *commit = pop_commit(&todo);
 
-		if (commit->object.flags & (UNINTERESTING | COUNTED))
-			break;
-		if (!(commit->object.flags & TREESAME))
-			nr++;
-		commit->object.flags |= COUNTED;
-		p = commit->parents;
-		entry = p;
-		if (p) {
-			p = p->next;
-			while (p) {
-				nr += count_distance(p);
-				p = p->next;
+		if (!(commit->object.flags & (UNINTERESTING | COUNTED))) {
+			struct commit_list *p;
+			if (!(commit->object.flags & TREESAME))
+				nr++;
+			commit->object.flags |= COUNTED;
+
+			for (p = commit->parents; p; p = p->next) {
+				commit_list_insert(p->item, &todo);
 			}
 		}
 	}
@@ -287,7 +277,7 @@ static struct commit_list *do_find_bisection(struct commit_list *list,
 	 * can reach.  So we do not have to run the expensive
 	 * count_distance() for single strand of pearls.
 	 *
-	 * However, if you have more than one parents, you cannot
+	 * However, if you have more than one parent, you cannot
 	 * just add their distance and one for yourself, since
 	 * they usually reach the same ancestor and you would
 	 * end up counting them twice that way.
@@ -296,17 +286,16 @@ static struct commit_list *do_find_bisection(struct commit_list *list,
 	 * way, and then fill the blanks using cheaper algorithm.
 	 */
 	for (p = list; p; p = p->next) {
-		if (p->item->object.flags & UNINTERESTING)
-			continue;
-		if (weight(p) != -2)
-			continue;
-		weight_set(p, count_distance(p));
-		clear_distance(list);
+		if (!(p->item->object.flags & UNINTERESTING)
+		 && (weight(p) == -2)) {
+			weight_set(p, count_distance(p));
+			clear_distance(list);
 
-		/* Does it happen to be at exactly half-way? */
-		if (!find_all && halfway(p, nr))
-			return p;
-		counted++;
+			/* Does it happen to be at exactly half-way? */
+			if (!find_all && halfway(p, nr))
+				return p;
+			counted++;
+		}
 	}
 
 	show_list("bisection 2 count_distance", counted, nr, list);
-- 
2.7.1.354.gd492730.dirty

  parent reply	other threads:[~2016-02-26  2:06 UTC|newest]

Thread overview: 34+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2016-02-26  2:04 [PATCH 00/16] git bisect improvements Stephan Beyer
2016-02-26  2:04 ` [PATCH 01/16] bisect: write about `bisect next` in documentation Stephan Beyer
2016-02-26  8:02   ` Jacob Keller
2016-02-26 18:47   ` Junio C Hamano
2016-02-27 13:45     ` Stephan Beyer
2016-02-27 18:03       ` Junio C Hamano
2016-02-27 19:38         ` Stephan Beyer
2016-02-28 18:28           ` Junio C Hamano
2016-02-26  2:04 ` [PATCH 02/16] bisect: add test for the bisect algorithm Stephan Beyer
2016-02-26  6:53   ` Christian Couder
2016-02-26 21:38     ` Stephan Beyer
2016-02-27 11:40       ` Christian Couder
2016-02-27 12:42         ` Matthieu Moy
2016-02-26  2:04 ` [PATCH 03/16] bisect: make bisect compile if DEBUG_BISECT is set Stephan Beyer
2016-02-26  2:04 ` [PATCH 04/16] bisect: make algorithm behavior independent of DEBUG_BISECT Stephan Beyer
2016-02-26  2:04 ` Stephan Beyer [this message]
2016-02-26  2:04 ` [PATCH 06/16] bisect: use struct node_data array instead of int array Stephan Beyer
2016-02-26  2:04 ` [PATCH 07/16] bisect: replace clear_distance() by unique markers Stephan Beyer
2016-02-26  2:04 ` [PATCH 08/16] bisect: use commit instead of commit list as arguments when appropriate Stephan Beyer
2016-02-26  3:10   ` Junio C Hamano
2016-02-26  2:04 ` [PATCH 09/16] bisect: extract get_distance() function from code duplication Stephan Beyer
2016-02-26  2:04 ` [PATCH 10/16] bisect: introduce distance_direction() Stephan Beyer
2016-02-26  2:04 ` [PATCH 11/16] bisect: make total number of commits global Stephan Beyer
2016-02-26  2:04 ` [PATCH 12/16] bisect: rename count_distance() to compute_weight() Stephan Beyer
2016-02-26  2:04 ` [PATCH 13/16] bisect: prepare for different algorithms based on find_all Stephan Beyer
2016-02-26  2:04 ` [PATCH 14/16] bisect: use a modified breadth-first search to find relevant weights Stephan Beyer
2016-02-26  3:09   ` Junio C Hamano
2016-02-26 20:55     ` Stephan Beyer
2016-02-26  2:04 ` [PATCH 15/16] bisect: compute best bisection in compute_relevant_weights() Stephan Beyer
2016-02-26  2:04 ` [PATCH 16/16] bisect: get back halfway shortcut Stephan Beyer
2016-03-20 18:50 ` [PATCH 00/16] git bisect improvements Pranit Bauva
2016-03-21 22:22   ` Stephan Beyer
2016-03-22  7:35     ` Christian Couder
2016-03-22 11:35     ` Pranit Bauva

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1456452282-10325-6-git-send-email-s-beyer@gmx.net \
    --to=s-beyer@gmx.net \
    --cc=chriscool@tuxfamily.org \
    --cc=git@vger.kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).