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

[PATCH 24/27] bisect: Stop bisection when we are confident about bad commit

From
JKJan Kara <jack@suse.cz>
Date
Nov 18, 2021, 16:49 UTC
Message-ID
<20211118164940.8818-25-jack@suse.cz>
In-Reply-To
<20211118164940.8818-1-jack@suse.cz>

When we found a commit that has high enough probability of being bad, stop bisection and report it.

Signed-off-by: Jan Kara <jack@suse.cz>
---
 bisect.c | 18 +++++++++++++++---
 1 file changed, 15 insertions(+), 3 deletions(-)
diff --git a/bisect.c b/bisect.c
index 12b027b86e75..7da74778d780 100644
--- a/bisect.c
+++ b/bisect.c
@@ -624,7 +624,7 @@ static int sw_rev_bmp_cmp(const void *data, const struct hashmap_entry *ap,
  * Compute for each commit a probability it is the bad one given tests
  * performed so far.
  */
-static void compute_commit_weights(struct commit_list *list)
+static struct commit_list *compute_commit_weights(struct commit_list *list)
 {
 	struct commit_list *p;
 	struct hashmap reach_map;
@@ -714,9 +714,14 @@ static void compute_commit_weights(struct commit_list *list)
 		found_entry = hashmap_get_entry(&reach_map, &entry, entry, NULL);
 		pweight->node_weight =
 			found_entry->cluster_p_bad / found_entry->count;
+		/* Found node we are confident enough is bad? */
+		if (pweight->node_weight >= result_confidence)
+			break;
 	}
 
 	hashmap_clear_and_free(&reach_map, struct sw_rev_bmp_hash_entry, entry);
+
+	return p;
 }
 
 void find_bisection(struct commit_list **commit_list, int *reaches,
@@ -758,14 +763,21 @@ void find_bisection(struct commit_list **commit_list, int *reaches,
 	if (result_confidence)
 		compute_tested_descendants(list);
 	list = reverse_list(list);
-	if (result_confidence)
-		compute_commit_weights(list);
+	if (result_confidence) {
+		best = compute_commit_weights(list);
+		/* Found commit we are confident is bad? Stop bisection... */
+		if (best) {
+			oidcpy(current_bad_oid, &best->item->object.oid);
+			goto found_best;
+		}
+	}
 	show_list("bisection 2 sorted", 0, nr, list);
 
 	/* Do the real work of finding bisection commit. */
 	best = do_find_bisection(list, nr, bisect_flags);
 	if (best) {
 		if (!(bisect_flags & FIND_BISECTION_ALL)) {
+found_best:
 			list->item = best->item;
 			free_commit_list(list->next);
 			best = list;
-- 
2.26.2
Previous: Jan KaraNext: Jan Kara
Message 33 of 43 in “Stochastic bisection support”
  1. Jan KaraNov 18, 2021
  2. 04/27 bisect: Fixup bisect-porcelain/32Jan Kara, Nov 18, 2021
  3. 02/27 bisect: Fixup bisect-porcelain/17Jan Kara, Nov 18, 2021
  4. Taylor BlauNov 18, 2021
  5. Jan KaraNov 22, 2021
  6. 03/27 bisect: Fixup test bisect-porcelain/20Jan Kara, Nov 18, 2021
  7. Chris TorekNov 18, 2021
  8. Taylor BlauNov 18, 2021
  9. Jan KaraNov 22, 2021
  10. 01/27 bisect: Fixup test rev-list-bisect/02Jan Kara, Nov 18, 2021
  11. Chris TorekNov 18, 2021
  12. Johannes SchindelinNov 19, 2021
  13. Jan KaraNov 22, 2021
  14. 10/27 bisect: Fixup bisect-porcelain/58Jan Kara, Nov 18, 2021
  15. 08/27 bisect: Fixup bisect-porcelain/50Jan Kara, Nov 18, 2021
  16. 05/27 bisect: Fixup bisect-porcelain/34Jan Kara, Nov 18, 2021
  17. 09/27 bisect: Fixup bisect-porcelain/54Jan Kara, Nov 18, 2021
  18. 06/27 bisect: Fixup bisect-porcelain/40Jan Kara, Nov 18, 2021
  19. 15/27 bisect: Rename clear_distance() to clear_counted_flag()Jan Kara, Nov 18, 2021
  20. 20/27 bisect: Compute probability a particular commit is badJan Kara, Nov 18, 2021
  21. 23/27 bisect: Find bisection point for stochastic weightsJan Kara, Nov 18, 2021
  22. 11/27 bisect: Fix bisection debuggingJan Kara, Nov 18, 2021
  23. 13/27 bisect: Allow specifying desired result confidenceJan Kara, Nov 18, 2021
  24. 22/27 bisect: Move count_distance()Jan Kara, Nov 18, 2021
  25. 19/27 bisect: Compute reachability of tested revsJan Kara, Nov 18, 2021
  26. 07/27 bisect: Remove duplicated bisect-porcelain/48Jan Kara, Nov 18, 2021
  27. 16/27 bisect: Separate commit list reversalJan Kara, Nov 18, 2021
  28. 14/27 bisect: Use void * for commit_weightJan Kara, Nov 18, 2021
  29. 17/27 bisect: Allow more complex commit weightsJan Kara, Nov 18, 2021
  30. 18/27 bisect: Terminate early if there are no eligible commitsJan Kara, Nov 18, 2021
  31. 21/27 bisect: Reorganize commit weight computationJan Kara, Nov 18, 2021
  32. 12/27 bisect: Accept and store confidence with each decisionJan Kara, Nov 18, 2021
  33. 24/27 bisect: Stop bisection when we are confident about bad commitJan Kara, Nov 18, 2021
  34. 26/27 bisect: Debug stochastic bisectionJan Kara, Nov 18, 2021
  35. 25/27 bisect: Report commit with the highest probabilityJan Kara, Nov 18, 2021
  36. 27/27 bisect: Allow bisection debugging of approx_halfway()Jan Kara, Nov 18, 2021
  37. Taylor BlauNov 18, 2021
  38. Jan KaraNov 22, 2021
  39. Johannes SchindelinNov 19, 2021
  40. Chris TorekNov 20, 2021
  41. Jan KaraNov 22, 2021
  42. Christian CouderNov 22, 2021
  43. Jan KaraNov 22, 2021

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.