freshlybakedca.ke
Git fork
commit: convert pop_most_recent_commit() to prio_queue
pop_most_recent_commit() calls commit_list_insert_by_date() for parent
commits, which is itself called in a loop. This can lead to quadratic
complexity if there are many merges. Replace the commit_list with a
prio_queue to ensure logarithmic worst case complexity and convert all
three users.
Add a performance test that exercises one of them using a pathological
history that consists of 50% merges and 50% root commits to demonstrate
the speedup:
Test v2.50.1 HEAD
----------------------------------------------------------------------
1501.2: rev-parse ':/65535' 2.48(2.47+0.00) 0.20(0.19+0.00) -91.9%
Alas, sane histories don't benefit from the conversion much, and
traversing Git's own history takes a 1% performance hit on my machine:
$ hyperfine -w3 -L git ./git_2.50.1,./git '{git} rev-parse :/^Initial.revision'
Benchmark 1: ./git_2.50.1 rev-parse :/^Initial.revision
Time (mean ± σ): 1.071 s ± 0.004 s [User: 1.052 s, System: 0.017 s]
Range (min … max): 1.067 s … 1.078 s 10 runs
Benchmark 2: ./git rev-parse :/^Initial.revision
Time (mean ± σ): 1.079 s ± 0.003 s [User: 1.060 s, System: 0.017 s]
Range (min … max): 1.074 s … 1.083 s 10 runs
Summary
./git_2.50.1 rev-parse :/^Initial.revision ran
1.01 ± 0.00 times faster than ./git rev-parse :/^Initial.revision
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
authored by
René Scharfe
and committed by
Junio C Hamano
d6ec0878
16bd9f20
+4
-3
commit.c
+4
-3
commit.c
···
31
31
#include "parse.h"
32
32
#include "object-file.h"
33
33
#include "object-file-convert.h"
34
+
#include "prio-queue.h"
34
35
35
36
static struct commit_extra_header *read_commit_extra_header_lines(const char *buf, size_t len, const char **);
36
37
···
738
739
commit_list_sort(list, commit_list_compare_by_date);
739
740
}
740
741
741
-
struct commit *pop_most_recent_commit(struct commit_list **list,
742
+
struct commit *pop_most_recent_commit(struct prio_queue *queue,
742
743
unsigned int mark)
743
744
{
744
-
struct commit *ret = pop_commit(list);
745
+
struct commit *ret = prio_queue_get(queue);
745
746
struct commit_list *parents = ret->parents;
746
747
747
748
while (parents) {
748
749
struct commit *commit = parents->item;
749
750
if (!repo_parse_commit(the_repository, commit) && !(commit->object.flags & mark)) {
750
751
commit->object.flags |= mark;
751
-
commit_list_insert_by_date(commit, list);
752
+
prio_queue_put(queue, commit);
752
753
}
753
754
parents = parents->next;
754
755
}
+4
-4
commit.h
+4
-4
commit.h
···
201
201
202
202
const char *skip_blank_lines(const char *msg);
203
203
204
-
/** Removes the first commit from a list sorted by date, and adds all
205
-
* of its parents.
206
-
**/
207
-
struct commit *pop_most_recent_commit(struct commit_list **list,
204
+
struct prio_queue;
205
+
206
+
/* Removes the first commit from a prio_queue and adds its parents. */
207
+
struct commit *pop_most_recent_commit(struct prio_queue *queue,
208
208
unsigned int mark);
209
209
210
210
struct commit *pop_commit(struct commit_list **stack);
+8
-5
fetch-pack.c
+8
-5
fetch-pack.c
···
34
34
#include "commit-graph.h"
35
35
#include "sigchain.h"
36
36
#include "mergesort.h"
37
+
#include "prio-queue.h"
37
38
38
39
static int transfer_unpack_limit = -1;
39
40
static int fetch_unpack_limit = -1;
···
600
601
return count ? retval : 0;
601
602
}
602
603
603
-
static struct commit_list *complete;
604
+
static struct prio_queue complete = { compare_commits_by_commit_date };
604
605
605
606
static int mark_complete(const struct object_id *oid)
606
607
{
···
608
609
609
610
if (commit && !(commit->object.flags & COMPLETE)) {
610
611
commit->object.flags |= COMPLETE;
611
-
commit_list_insert(commit, &complete);
612
+
prio_queue_put(&complete, commit);
612
613
}
613
614
return 0;
614
615
}
···
625
626
static void mark_recent_complete_commits(struct fetch_pack_args *args,
626
627
timestamp_t cutoff)
627
628
{
628
-
while (complete && cutoff <= complete->item->date) {
629
+
while (complete.nr) {
630
+
struct commit *item = prio_queue_peek(&complete);
631
+
if (item->date < cutoff)
632
+
break;
629
633
print_verbose(args, _("Marking %s as complete"),
630
-
oid_to_hex(&complete->item->object.oid));
634
+
oid_to_hex(&item->object.oid));
631
635
pop_most_recent_commit(&complete, COMPLETE);
632
636
}
633
637
}
···
797
801
refs_for_each_rawref(get_main_ref_store(the_repository),
798
802
mark_complete_oid, NULL);
799
803
for_each_cached_alternate(NULL, mark_alternate_complete);
800
-
commit_list_sort_by_date(&complete);
801
804
if (cutoff)
802
805
mark_recent_complete_commits(args, cutoff);
803
806
}
+5
-5
object-name.c
+5
-5
object-name.c
···
28
28
#include "commit-reach.h"
29
29
#include "date.h"
30
30
#include "object-file-convert.h"
31
+
#include "prio-queue.h"
31
32
32
33
static int get_oid_oneline(struct repository *r, const char *, struct object_id *,
33
34
const struct commit_list *);
···
1457
1458
const char *prefix, struct object_id *oid,
1458
1459
const struct commit_list *list)
1459
1460
{
1460
-
struct commit_list *copy = NULL, **copy_tail = ©
1461
+
struct prio_queue copy = { compare_commits_by_commit_date };
1461
1462
const struct commit_list *l;
1462
1463
int found = 0;
1463
1464
int negative = 0;
···
1479
1480
1480
1481
for (l = list; l; l = l->next) {
1481
1482
l->item->object.flags |= ONELINE_SEEN;
1482
-
copy_tail = &commit_list_insert(l->item, copy_tail)->next;
1483
+
prio_queue_put(©, l->item);
1483
1484
}
1484
-
while (copy) {
1485
+
while (copy.nr) {
1485
1486
const char *p, *buf;
1486
1487
struct commit *commit;
1487
1488
int matches;
···
1503
1504
regfree(®ex);
1504
1505
for (l = list; l; l = l->next)
1505
1506
clear_commit_marks(l->item, ONELINE_SEEN);
1506
-
free_commit_list(copy);
1507
+
clear_prio_queue(©);
1507
1508
return found ? 0 : -1;
1508
1509
}
1509
1510
···
2057
2058
cb.list = &list;
2058
2059
refs_for_each_ref(get_main_ref_store(repo), handle_one_ref, &cb);
2059
2060
refs_head_ref(get_main_ref_store(repo), handle_one_ref, &cb);
2060
-
commit_list_sort_by_date(&list);
2061
2061
ret = get_oid_oneline(repo, name + 2, oid, list);
2062
2062
2063
2063
free_commit_list(list);
+1
t/meson.build
+1
t/meson.build
+71
t/perf/p1501-rev-parse-oneline.sh
+71
t/perf/p1501-rev-parse-oneline.sh
···
1
+
#!/bin/sh
2
+
3
+
test_description='Test :/ object name notation'
4
+
5
+
. ./perf-lib.sh
6
+
7
+
test_perf_fresh_repo
8
+
9
+
#
10
+
# Creates lots of merges to make history traversal costly. In
11
+
# particular it creates 2^($max_level-1)-1 2-way merges on top of
12
+
# 2^($max_level-1) root commits. E.g., the commit history looks like
13
+
# this for a $max_level of 3:
14
+
#
15
+
# _1_
16
+
# / \
17
+
# 2 3
18
+
# / \ / \
19
+
# 4 5 6 7
20
+
#
21
+
# The numbers are the fast-import marks, which also are the commit
22
+
# messages. 1 is the HEAD commit and a merge, 2 and 3 are also merges,
23
+
# 4-7 are the root commits.
24
+
#
25
+
build_history () {
26
+
local max_level="$1" &&
27
+
local level="${2:-1}" &&
28
+
local mark="${3:-1}" &&
29
+
if test $level -eq $max_level
30
+
then
31
+
echo "reset refs/heads/master" &&
32
+
echo "from $ZERO_OID" &&
33
+
echo "commit refs/heads/master" &&
34
+
echo "mark :$mark" &&
35
+
echo "committer C <c@example.com> 1234567890 +0000" &&
36
+
echo "data <<EOF" &&
37
+
echo "$mark" &&
38
+
echo "EOF"
39
+
else
40
+
local level1=$((level+1)) &&
41
+
local mark1=$((2*mark)) &&
42
+
local mark2=$((2*mark+1)) &&
43
+
build_history $max_level $level1 $mark1 &&
44
+
build_history $max_level $level1 $mark2 &&
45
+
echo "commit refs/heads/master" &&
46
+
echo "mark :$mark" &&
47
+
echo "committer C <c@example.com> 1234567890 +0000" &&
48
+
echo "data <<EOF" &&
49
+
echo "$mark" &&
50
+
echo "EOF" &&
51
+
echo "from :$mark1" &&
52
+
echo "merge :$mark2"
53
+
fi
54
+
}
55
+
56
+
test_expect_success 'setup' '
57
+
build_history 16 | git fast-import &&
58
+
git log --format="%H %s" --reverse >commits &&
59
+
sed -n -e "s/ .*$//p" -e "q" <commits >expect &&
60
+
sed -n -e "s/^.* //p" -e "q" <commits >needle
61
+
'
62
+
63
+
test_perf "rev-parse :/$(cat needle)" '
64
+
git rev-parse :/$(cat needle) >actual
65
+
'
66
+
67
+
test_expect_success 'verify result' '
68
+
test_cmp expect actual
69
+
'
70
+
71
+
test_done
+7
-4
walker.c
+7
-4
walker.c
···
14
14
#include "blob.h"
15
15
#include "refs.h"
16
16
#include "progress.h"
17
+
#include "prio-queue.h"
17
18
18
19
static struct object_id current_commit_oid;
19
20
···
78
79
#define SEEN (1U << 1)
79
80
#define TO_SCAN (1U << 2)
80
81
81
-
static struct commit_list *complete = NULL;
82
+
static struct prio_queue complete = { compare_commits_by_commit_date };
82
83
83
84
static int process_commit(struct walker *walker, struct commit *commit)
84
85
{
···
87
88
if (repo_parse_commit(the_repository, commit))
88
89
return -1;
89
90
90
-
while (complete && complete->item->date >= commit->date) {
91
+
while (complete.nr) {
92
+
struct commit *item = prio_queue_peek(&complete);
93
+
if (item->date < commit->date)
94
+
break;
91
95
pop_most_recent_commit(&complete, COMPLETE);
92
96
}
93
97
···
233
237
234
238
if (commit) {
235
239
commit->object.flags |= COMPLETE;
236
-
commit_list_insert(commit, &complete);
240
+
prio_queue_put(&complete, commit);
237
241
}
238
242
return 0;
239
243
}
···
302
306
if (!walker->get_recover) {
303
307
refs_for_each_ref(get_main_ref_store(the_repository),
304
308
mark_complete, NULL);
305
-
commit_list_sort_by_date(&complete);
306
309
}
307
310
308
311
for (i = 0; i < targets; i++) {