global: introduce `USE_THE_REPOSITORY_VARIABLE` macro
Use of the `the_repository` variable is deprecated nowadays, and we
slowly but steadily convert the codebase to not use it anymore. Instead,
callers should be passing down the repository to work on via parameters.
It is hard though to prove that a given code unit does not use this
variable anymore. The most trivial case, merely demonstrating that there
is no direct use of `the_repository`, is already a bit of a pain during
code reviews as the reviewer needs to manually verify claims made by the
patch author. The bigger problem though is that we have many interfaces
that implicitly rely on `the_repository`.
Introduce a new `USE_THE_REPOSITORY_VARIABLE` macro that allows code
units to opt into usage of `the_repository`. The intent of this macro is
to demonstrate that a certain code unit does not use this variable
anymore, and to keep it from new dependencies on it in future changes,
be it explicit or implicit
For now, the macro only guards `the_repository` itself as well as
`the_hash_algo`. There are many more known interfaces where we have an
implicit dependency on `the_repository`, but those are not guarded at
the current point in time. Over time though, we should start to add
guards as required (or even better, just remove them).
Define the macro as required in our code units. As expected, most of our
code still relies on the global variable. Nearly all of our builtins
rely on the variable as there is no way yet to pass `the_repository` to
their entry point. For now, declare the macro in "biultin.h" to keep the
required changes at least a little bit more contained.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2024-06-14 14:50:23 +08:00
|
|
|
#define USE_THE_REPOSITORY_VARIABLE
|
|
|
|
|
2023-04-11 15:41:56 +08:00
|
|
|
#include "git-compat-util.h"
|
2017-06-15 02:07:36 +08:00
|
|
|
#include "config.h"
|
2009-03-26 12:55:24 +08:00
|
|
|
#include "commit.h"
|
|
|
|
#include "diff.h"
|
2023-03-21 14:26:03 +08:00
|
|
|
#include "environment.h"
|
2023-03-21 14:25:54 +08:00
|
|
|
#include "gettext.h"
|
2023-02-24 08:09:27 +08:00
|
|
|
#include "hex.h"
|
2009-03-26 12:55:24 +08:00
|
|
|
#include "revision.h"
|
2009-03-26 12:55:54 +08:00
|
|
|
#include "refs.h"
|
|
|
|
#include "list-objects.h"
|
2009-03-26 12:55:59 +08:00
|
|
|
#include "quote.h"
|
2009-04-19 17:56:07 +08:00
|
|
|
#include "run-command.h"
|
2009-05-29 05:21:16 +08:00
|
|
|
#include "log-tree.h"
|
2009-03-26 12:55:24 +08:00
|
|
|
#include "bisect.h"
|
2020-03-30 22:03:46 +08:00
|
|
|
#include "oid-array.h"
|
2020-07-29 04:23:39 +08:00
|
|
|
#include "strvec.h"
|
2018-05-19 13:28:25 +08:00
|
|
|
#include "commit-slab.h"
|
2018-07-21 00:33:04 +08:00
|
|
|
#include "commit-reach.h"
|
2023-04-11 15:41:49 +08:00
|
|
|
#include "object-name.h"
|
2023-05-16 14:34:06 +08:00
|
|
|
#include "object-store-ll.h"
|
2023-05-16 14:33:59 +08:00
|
|
|
#include "path.h"
|
2020-08-08 05:58:37 +08:00
|
|
|
#include "dir.h"
|
2009-05-09 23:55:38 +08:00
|
|
|
|
2017-03-31 09:40:00 +08:00
|
|
|
static struct oid_array good_revs;
|
|
|
|
static struct oid_array skipped_revs;
|
2009-03-26 12:55:54 +08:00
|
|
|
|
2015-03-14 07:39:29 +08:00
|
|
|
static struct object_id *current_bad_oid;
|
2009-04-19 17:56:07 +08:00
|
|
|
|
2015-06-29 23:40:29 +08:00
|
|
|
static const char *term_bad;
|
|
|
|
static const char *term_good;
|
|
|
|
|
2014-03-25 21:23:26 +08:00
|
|
|
/* Remember to update object flag allocation in object.h */
|
2009-03-26 12:55:24 +08:00
|
|
|
#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;
|
|
|
|
|
|
|
|
while (entry) {
|
|
|
|
struct commit *commit = entry->item;
|
|
|
|
struct commit_list *p;
|
|
|
|
|
|
|
|
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;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return nr;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void clear_distance(struct commit_list *list)
|
|
|
|
{
|
|
|
|
while (list) {
|
|
|
|
struct commit *commit = list->item;
|
|
|
|
commit->object.flags &= ~COUNTED;
|
|
|
|
list = list->next;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-05-19 13:28:25 +08:00
|
|
|
define_commit_slab(commit_weight, int *);
|
|
|
|
static struct commit_weight commit_weight;
|
|
|
|
|
2009-03-26 12:55:24 +08:00
|
|
|
#define DEBUG_BISECT 0
|
|
|
|
|
|
|
|
static inline int weight(struct commit_list *elem)
|
|
|
|
{
|
2018-05-19 13:28:25 +08:00
|
|
|
return **commit_weight_at(&commit_weight, elem->item);
|
2009-03-26 12:55:24 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static inline void weight_set(struct commit_list *elem, int weight)
|
|
|
|
{
|
2018-05-19 13:28:25 +08:00
|
|
|
**commit_weight_at(&commit_weight, elem->item) = weight;
|
2009-03-26 12:55:24 +08:00
|
|
|
}
|
|
|
|
|
2020-08-08 05:58:38 +08:00
|
|
|
static int count_interesting_parents(struct commit *commit, unsigned bisect_flags)
|
2009-03-26 12:55:24 +08:00
|
|
|
{
|
|
|
|
struct commit_list *p;
|
|
|
|
int count;
|
|
|
|
|
|
|
|
for (count = 0, p = commit->parents; p; p = p->next) {
|
2020-08-08 05:58:35 +08:00
|
|
|
if (!(p->item->object.flags & UNINTERESTING))
|
|
|
|
count++;
|
2020-08-08 05:58:38 +08:00
|
|
|
if (bisect_flags & FIND_BISECTION_FIRST_PARENT_ONLY)
|
2020-08-08 05:58:35 +08:00
|
|
|
break;
|
2009-03-26 12:55:24 +08:00
|
|
|
}
|
|
|
|
return count;
|
|
|
|
}
|
|
|
|
|
bisect: loosen halfway() check for a large number of commits
'git bisect start ...' and subsequent 'git bisect (good|bad)' commands
can take quite a while when the given/remaining revision range between
good and bad commits is big and contains a lot of merge commits, e.g.
in git.git:
$ git rev-list --count v1.6.0..v2.28.0
44284
$ time git bisect start v2.28.0 v1.6.0
Bisecting: 22141 revisions left to test after this (roughly 15 steps)
[e197c21807dacadc8305250baa0b9228819189d4] unable_to_lock_die(): rename function from unable_to_lock_index_die()
real 0m15.472s
user 0m15.220s
sys 0m0.255s
The majority of the runtime is spent in do_find_bisection(), where we
try to find a commit as close as possible to the halfway point between
the bad and good revisions, i.e. a commit from which the number of
reachable commits that are in the good-bad range is half the total
number of commits in that range. So we count how many commits are
reachable in the good-bad range for each commit in that range, which
is quick and easy for a linear history, even over 300k commits in a
linear range are handled in ~0.3s on my machine. Alas, handling merge
commits is non-trivial and quite expensive as the algorithm used seems
to be quadratic, causing the long runtime shown above.
Interestingly, look at what a big difference one additional commit
can make:
$ git rev-list --count v1.6.0^..v2.28.0
44285
$ time git bisect start v2.28.0 v1.6.0^
Bisecting: 22142 revisions left to test after this (roughly 15 steps)
[565301e41670825ceedf75220f2918ae76831240] Sync with 2.1.2
real 0m5.848s
user 0m5.600s
sys 0m0.252s
The difference is caused by one of the optimizations attempting to cut
down the runtime added in 1c4fea3a40 (git-rev-list --bisect:
optimization, 2007-03-21):
Another small optimization is whenever we find a half-way commit
(that is, a commit that can reach exactly half of the commits),
we stop giving counts to remaining commits, as we will not find
any better commit than we just found.
In this second 'git bisect start' command we happen to find a commit
exactly at the halfway point and can return early, but in the first
case there is no such commit, so we can't return early and end up
counting the number of reachable commits from all commits in the
good-bad range.
However, when we have thousands of commits it's not all that important
to find the _exact_ halfway point, a few commits more or less doesn't
make any real difference for the bisection.
So let's loosen the check in the halfway() helper to consider commits
within about 0.1% of the exact halfway point as halfway as well, and
rename the function to approx_halfway() accordingly. This will allow
us to return early on a bigger good-bad range, even when there is no
commit exactly at the halfway point, thereby reducing the runtime of
the first command above considerably, from ~15s to 4.901s.
Furthermore, even if there is a commit exactly at the halfway point,
we might still stumble upon a commit within that 0.1% range before
finding the exact halfway point, allowing us to return a bit earlier,
slightly reducing the runtime of the second command from 5.848s to
5.058s. Note that this change doesn't affect good-bad ranges
containing ~2000 commits or less, because that 0.1% tolerance becomes
zero due to integer arithmetic; however, if the range is that small
then counting the reachable commits for all commits is already fast
enough anyway.
Naturally, this will likely change which commits get picked at each
bisection step, and, in turn, might change how many bisection steps
are necessary to find the first bad commit. If the number of
necessary bisection steps were to increase often, then this change
could backfire, because building and testing at each step might take
much longer than the time spared. OTOH, if the number of steps were
to decrease, then it would be a double win.
So I ran some tests to see how often that happens: picked random good
and bad starting revisions at least 50k commits apart and a random
first bad commit in between in git.git, and used 'git bisect run git
merge-base --is-ancestor HEAD $first_bad_commit' to check the number
of necessary bisection steps. After repeating all this 1000 times
both with and without this patch I found that:
- 146 cases needed one more bisection step than before, 149 cases
needed one less step, while in the remaining 705 cases the number
of steps didn't change. So the number of bisection steps does
indeed change in a non-negligible number of cases, but it seems
that the average number of steps doesn't change in the long run.
- The first 'git bisect start' command got over 3x faster in 456
cases, so this "no commit at the exact halfway point" case seems
to be common enough to care about.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-11-13 00:19:38 +08:00
|
|
|
static inline int approx_halfway(struct commit_list *p, int nr)
|
2009-03-26 12:55:24 +08:00
|
|
|
{
|
bisect: loosen halfway() check for a large number of commits
'git bisect start ...' and subsequent 'git bisect (good|bad)' commands
can take quite a while when the given/remaining revision range between
good and bad commits is big and contains a lot of merge commits, e.g.
in git.git:
$ git rev-list --count v1.6.0..v2.28.0
44284
$ time git bisect start v2.28.0 v1.6.0
Bisecting: 22141 revisions left to test after this (roughly 15 steps)
[e197c21807dacadc8305250baa0b9228819189d4] unable_to_lock_die(): rename function from unable_to_lock_index_die()
real 0m15.472s
user 0m15.220s
sys 0m0.255s
The majority of the runtime is spent in do_find_bisection(), where we
try to find a commit as close as possible to the halfway point between
the bad and good revisions, i.e. a commit from which the number of
reachable commits that are in the good-bad range is half the total
number of commits in that range. So we count how many commits are
reachable in the good-bad range for each commit in that range, which
is quick and easy for a linear history, even over 300k commits in a
linear range are handled in ~0.3s on my machine. Alas, handling merge
commits is non-trivial and quite expensive as the algorithm used seems
to be quadratic, causing the long runtime shown above.
Interestingly, look at what a big difference one additional commit
can make:
$ git rev-list --count v1.6.0^..v2.28.0
44285
$ time git bisect start v2.28.0 v1.6.0^
Bisecting: 22142 revisions left to test after this (roughly 15 steps)
[565301e41670825ceedf75220f2918ae76831240] Sync with 2.1.2
real 0m5.848s
user 0m5.600s
sys 0m0.252s
The difference is caused by one of the optimizations attempting to cut
down the runtime added in 1c4fea3a40 (git-rev-list --bisect:
optimization, 2007-03-21):
Another small optimization is whenever we find a half-way commit
(that is, a commit that can reach exactly half of the commits),
we stop giving counts to remaining commits, as we will not find
any better commit than we just found.
In this second 'git bisect start' command we happen to find a commit
exactly at the halfway point and can return early, but in the first
case there is no such commit, so we can't return early and end up
counting the number of reachable commits from all commits in the
good-bad range.
However, when we have thousands of commits it's not all that important
to find the _exact_ halfway point, a few commits more or less doesn't
make any real difference for the bisection.
So let's loosen the check in the halfway() helper to consider commits
within about 0.1% of the exact halfway point as halfway as well, and
rename the function to approx_halfway() accordingly. This will allow
us to return early on a bigger good-bad range, even when there is no
commit exactly at the halfway point, thereby reducing the runtime of
the first command above considerably, from ~15s to 4.901s.
Furthermore, even if there is a commit exactly at the halfway point,
we might still stumble upon a commit within that 0.1% range before
finding the exact halfway point, allowing us to return a bit earlier,
slightly reducing the runtime of the second command from 5.848s to
5.058s. Note that this change doesn't affect good-bad ranges
containing ~2000 commits or less, because that 0.1% tolerance becomes
zero due to integer arithmetic; however, if the range is that small
then counting the reachable commits for all commits is already fast
enough anyway.
Naturally, this will likely change which commits get picked at each
bisection step, and, in turn, might change how many bisection steps
are necessary to find the first bad commit. If the number of
necessary bisection steps were to increase often, then this change
could backfire, because building and testing at each step might take
much longer than the time spared. OTOH, if the number of steps were
to decrease, then it would be a double win.
So I ran some tests to see how often that happens: picked random good
and bad starting revisions at least 50k commits apart and a random
first bad commit in between in git.git, and used 'git bisect run git
merge-base --is-ancestor HEAD $first_bad_commit' to check the number
of necessary bisection steps. After repeating all this 1000 times
both with and without this patch I found that:
- 146 cases needed one more bisection step than before, 149 cases
needed one less step, while in the remaining 705 cases the number
of steps didn't change. So the number of bisection steps does
indeed change in a non-negligible number of cases, but it seems
that the average number of steps doesn't change in the long run.
- The first 'git bisect start' command got over 3x faster in 456
cases, so this "no commit at the exact halfway point" case seems
to be common enough to care about.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-11-13 00:19:38 +08:00
|
|
|
int diff;
|
|
|
|
|
2009-03-26 12:55:24 +08:00
|
|
|
/*
|
|
|
|
* Don't short-cut something we are not going to return!
|
|
|
|
*/
|
|
|
|
if (p->item->object.flags & TREESAME)
|
|
|
|
return 0;
|
|
|
|
if (DEBUG_BISECT)
|
|
|
|
return 0;
|
|
|
|
/*
|
bisect: loosen halfway() check for a large number of commits
'git bisect start ...' and subsequent 'git bisect (good|bad)' commands
can take quite a while when the given/remaining revision range between
good and bad commits is big and contains a lot of merge commits, e.g.
in git.git:
$ git rev-list --count v1.6.0..v2.28.0
44284
$ time git bisect start v2.28.0 v1.6.0
Bisecting: 22141 revisions left to test after this (roughly 15 steps)
[e197c21807dacadc8305250baa0b9228819189d4] unable_to_lock_die(): rename function from unable_to_lock_index_die()
real 0m15.472s
user 0m15.220s
sys 0m0.255s
The majority of the runtime is spent in do_find_bisection(), where we
try to find a commit as close as possible to the halfway point between
the bad and good revisions, i.e. a commit from which the number of
reachable commits that are in the good-bad range is half the total
number of commits in that range. So we count how many commits are
reachable in the good-bad range for each commit in that range, which
is quick and easy for a linear history, even over 300k commits in a
linear range are handled in ~0.3s on my machine. Alas, handling merge
commits is non-trivial and quite expensive as the algorithm used seems
to be quadratic, causing the long runtime shown above.
Interestingly, look at what a big difference one additional commit
can make:
$ git rev-list --count v1.6.0^..v2.28.0
44285
$ time git bisect start v2.28.0 v1.6.0^
Bisecting: 22142 revisions left to test after this (roughly 15 steps)
[565301e41670825ceedf75220f2918ae76831240] Sync with 2.1.2
real 0m5.848s
user 0m5.600s
sys 0m0.252s
The difference is caused by one of the optimizations attempting to cut
down the runtime added in 1c4fea3a40 (git-rev-list --bisect:
optimization, 2007-03-21):
Another small optimization is whenever we find a half-way commit
(that is, a commit that can reach exactly half of the commits),
we stop giving counts to remaining commits, as we will not find
any better commit than we just found.
In this second 'git bisect start' command we happen to find a commit
exactly at the halfway point and can return early, but in the first
case there is no such commit, so we can't return early and end up
counting the number of reachable commits from all commits in the
good-bad range.
However, when we have thousands of commits it's not all that important
to find the _exact_ halfway point, a few commits more or less doesn't
make any real difference for the bisection.
So let's loosen the check in the halfway() helper to consider commits
within about 0.1% of the exact halfway point as halfway as well, and
rename the function to approx_halfway() accordingly. This will allow
us to return early on a bigger good-bad range, even when there is no
commit exactly at the halfway point, thereby reducing the runtime of
the first command above considerably, from ~15s to 4.901s.
Furthermore, even if there is a commit exactly at the halfway point,
we might still stumble upon a commit within that 0.1% range before
finding the exact halfway point, allowing us to return a bit earlier,
slightly reducing the runtime of the second command from 5.848s to
5.058s. Note that this change doesn't affect good-bad ranges
containing ~2000 commits or less, because that 0.1% tolerance becomes
zero due to integer arithmetic; however, if the range is that small
then counting the reachable commits for all commits is already fast
enough anyway.
Naturally, this will likely change which commits get picked at each
bisection step, and, in turn, might change how many bisection steps
are necessary to find the first bad commit. If the number of
necessary bisection steps were to increase often, then this change
could backfire, because building and testing at each step might take
much longer than the time spared. OTOH, if the number of steps were
to decrease, then it would be a double win.
So I ran some tests to see how often that happens: picked random good
and bad starting revisions at least 50k commits apart and a random
first bad commit in between in git.git, and used 'git bisect run git
merge-base --is-ancestor HEAD $first_bad_commit' to check the number
of necessary bisection steps. After repeating all this 1000 times
both with and without this patch I found that:
- 146 cases needed one more bisection step than before, 149 cases
needed one less step, while in the remaining 705 cases the number
of steps didn't change. So the number of bisection steps does
indeed change in a non-negligible number of cases, but it seems
that the average number of steps doesn't change in the long run.
- The first 'git bisect start' command got over 3x faster in 456
cases, so this "no commit at the exact halfway point" case seems
to be common enough to care about.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-11-13 00:19:38 +08:00
|
|
|
* For small number of commits 2 and 3 are halfway of 5, and
|
2009-03-26 12:55:24 +08:00
|
|
|
* 3 is halfway of 6 but 2 and 4 are not.
|
|
|
|
*/
|
bisect: loosen halfway() check for a large number of commits
'git bisect start ...' and subsequent 'git bisect (good|bad)' commands
can take quite a while when the given/remaining revision range between
good and bad commits is big and contains a lot of merge commits, e.g.
in git.git:
$ git rev-list --count v1.6.0..v2.28.0
44284
$ time git bisect start v2.28.0 v1.6.0
Bisecting: 22141 revisions left to test after this (roughly 15 steps)
[e197c21807dacadc8305250baa0b9228819189d4] unable_to_lock_die(): rename function from unable_to_lock_index_die()
real 0m15.472s
user 0m15.220s
sys 0m0.255s
The majority of the runtime is spent in do_find_bisection(), where we
try to find a commit as close as possible to the halfway point between
the bad and good revisions, i.e. a commit from which the number of
reachable commits that are in the good-bad range is half the total
number of commits in that range. So we count how many commits are
reachable in the good-bad range for each commit in that range, which
is quick and easy for a linear history, even over 300k commits in a
linear range are handled in ~0.3s on my machine. Alas, handling merge
commits is non-trivial and quite expensive as the algorithm used seems
to be quadratic, causing the long runtime shown above.
Interestingly, look at what a big difference one additional commit
can make:
$ git rev-list --count v1.6.0^..v2.28.0
44285
$ time git bisect start v2.28.0 v1.6.0^
Bisecting: 22142 revisions left to test after this (roughly 15 steps)
[565301e41670825ceedf75220f2918ae76831240] Sync with 2.1.2
real 0m5.848s
user 0m5.600s
sys 0m0.252s
The difference is caused by one of the optimizations attempting to cut
down the runtime added in 1c4fea3a40 (git-rev-list --bisect:
optimization, 2007-03-21):
Another small optimization is whenever we find a half-way commit
(that is, a commit that can reach exactly half of the commits),
we stop giving counts to remaining commits, as we will not find
any better commit than we just found.
In this second 'git bisect start' command we happen to find a commit
exactly at the halfway point and can return early, but in the first
case there is no such commit, so we can't return early and end up
counting the number of reachable commits from all commits in the
good-bad range.
However, when we have thousands of commits it's not all that important
to find the _exact_ halfway point, a few commits more or less doesn't
make any real difference for the bisection.
So let's loosen the check in the halfway() helper to consider commits
within about 0.1% of the exact halfway point as halfway as well, and
rename the function to approx_halfway() accordingly. This will allow
us to return early on a bigger good-bad range, even when there is no
commit exactly at the halfway point, thereby reducing the runtime of
the first command above considerably, from ~15s to 4.901s.
Furthermore, even if there is a commit exactly at the halfway point,
we might still stumble upon a commit within that 0.1% range before
finding the exact halfway point, allowing us to return a bit earlier,
slightly reducing the runtime of the second command from 5.848s to
5.058s. Note that this change doesn't affect good-bad ranges
containing ~2000 commits or less, because that 0.1% tolerance becomes
zero due to integer arithmetic; however, if the range is that small
then counting the reachable commits for all commits is already fast
enough anyway.
Naturally, this will likely change which commits get picked at each
bisection step, and, in turn, might change how many bisection steps
are necessary to find the first bad commit. If the number of
necessary bisection steps were to increase often, then this change
could backfire, because building and testing at each step might take
much longer than the time spared. OTOH, if the number of steps were
to decrease, then it would be a double win.
So I ran some tests to see how often that happens: picked random good
and bad starting revisions at least 50k commits apart and a random
first bad commit in between in git.git, and used 'git bisect run git
merge-base --is-ancestor HEAD $first_bad_commit' to check the number
of necessary bisection steps. After repeating all this 1000 times
both with and without this patch I found that:
- 146 cases needed one more bisection step than before, 149 cases
needed one less step, while in the remaining 705 cases the number
of steps didn't change. So the number of bisection steps does
indeed change in a non-negligible number of cases, but it seems
that the average number of steps doesn't change in the long run.
- The first 'git bisect start' command got over 3x faster in 456
cases, so this "no commit at the exact halfway point" case seems
to be common enough to care about.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-11-13 00:19:38 +08:00
|
|
|
diff = 2 * weight(p) - nr;
|
|
|
|
switch (diff) {
|
2009-03-26 12:55:24 +08:00
|
|
|
case -1: case 0: case 1:
|
|
|
|
return 1;
|
|
|
|
default:
|
bisect: loosen halfway() check for a large number of commits
'git bisect start ...' and subsequent 'git bisect (good|bad)' commands
can take quite a while when the given/remaining revision range between
good and bad commits is big and contains a lot of merge commits, e.g.
in git.git:
$ git rev-list --count v1.6.0..v2.28.0
44284
$ time git bisect start v2.28.0 v1.6.0
Bisecting: 22141 revisions left to test after this (roughly 15 steps)
[e197c21807dacadc8305250baa0b9228819189d4] unable_to_lock_die(): rename function from unable_to_lock_index_die()
real 0m15.472s
user 0m15.220s
sys 0m0.255s
The majority of the runtime is spent in do_find_bisection(), where we
try to find a commit as close as possible to the halfway point between
the bad and good revisions, i.e. a commit from which the number of
reachable commits that are in the good-bad range is half the total
number of commits in that range. So we count how many commits are
reachable in the good-bad range for each commit in that range, which
is quick and easy for a linear history, even over 300k commits in a
linear range are handled in ~0.3s on my machine. Alas, handling merge
commits is non-trivial and quite expensive as the algorithm used seems
to be quadratic, causing the long runtime shown above.
Interestingly, look at what a big difference one additional commit
can make:
$ git rev-list --count v1.6.0^..v2.28.0
44285
$ time git bisect start v2.28.0 v1.6.0^
Bisecting: 22142 revisions left to test after this (roughly 15 steps)
[565301e41670825ceedf75220f2918ae76831240] Sync with 2.1.2
real 0m5.848s
user 0m5.600s
sys 0m0.252s
The difference is caused by one of the optimizations attempting to cut
down the runtime added in 1c4fea3a40 (git-rev-list --bisect:
optimization, 2007-03-21):
Another small optimization is whenever we find a half-way commit
(that is, a commit that can reach exactly half of the commits),
we stop giving counts to remaining commits, as we will not find
any better commit than we just found.
In this second 'git bisect start' command we happen to find a commit
exactly at the halfway point and can return early, but in the first
case there is no such commit, so we can't return early and end up
counting the number of reachable commits from all commits in the
good-bad range.
However, when we have thousands of commits it's not all that important
to find the _exact_ halfway point, a few commits more or less doesn't
make any real difference for the bisection.
So let's loosen the check in the halfway() helper to consider commits
within about 0.1% of the exact halfway point as halfway as well, and
rename the function to approx_halfway() accordingly. This will allow
us to return early on a bigger good-bad range, even when there is no
commit exactly at the halfway point, thereby reducing the runtime of
the first command above considerably, from ~15s to 4.901s.
Furthermore, even if there is a commit exactly at the halfway point,
we might still stumble upon a commit within that 0.1% range before
finding the exact halfway point, allowing us to return a bit earlier,
slightly reducing the runtime of the second command from 5.848s to
5.058s. Note that this change doesn't affect good-bad ranges
containing ~2000 commits or less, because that 0.1% tolerance becomes
zero due to integer arithmetic; however, if the range is that small
then counting the reachable commits for all commits is already fast
enough anyway.
Naturally, this will likely change which commits get picked at each
bisection step, and, in turn, might change how many bisection steps
are necessary to find the first bad commit. If the number of
necessary bisection steps were to increase often, then this change
could backfire, because building and testing at each step might take
much longer than the time spared. OTOH, if the number of steps were
to decrease, then it would be a double win.
So I ran some tests to see how often that happens: picked random good
and bad starting revisions at least 50k commits apart and a random
first bad commit in between in git.git, and used 'git bisect run git
merge-base --is-ancestor HEAD $first_bad_commit' to check the number
of necessary bisection steps. After repeating all this 1000 times
both with and without this patch I found that:
- 146 cases needed one more bisection step than before, 149 cases
needed one less step, while in the remaining 705 cases the number
of steps didn't change. So the number of bisection steps does
indeed change in a non-negligible number of cases, but it seems
that the average number of steps doesn't change in the long run.
- The first 'git bisect start' command got over 3x faster in 456
cases, so this "no commit at the exact halfway point" case seems
to be common enough to care about.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-11-13 00:19:38 +08:00
|
|
|
/*
|
|
|
|
* For large number of commits we are not so strict, it's
|
|
|
|
* good enough if it's within ~0.1% of the halfway point,
|
|
|
|
* e.g. 5000 is exactly halfway of 10000, but we consider
|
|
|
|
* the values [4996, 5004] as halfway as well.
|
|
|
|
*/
|
|
|
|
if (abs(diff) < nr / 1024)
|
|
|
|
return 1;
|
2009-03-26 12:55:24 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void show_list(const char *debug, int counted, int nr,
|
|
|
|
struct commit_list *list)
|
|
|
|
{
|
|
|
|
struct commit_list *p;
|
|
|
|
|
2018-09-02 15:42:50 +08:00
|
|
|
if (!DEBUG_BISECT)
|
|
|
|
return;
|
|
|
|
|
2009-03-26 12:55:24 +08:00
|
|
|
fprintf(stderr, "%s (%d/%d)\n", debug, counted, nr);
|
|
|
|
|
|
|
|
for (p = list; p; p = p->next) {
|
|
|
|
struct commit_list *pp;
|
|
|
|
struct commit *commit = p->item;
|
2020-08-08 05:58:38 +08:00
|
|
|
unsigned commit_flags = commit->object.flags;
|
2009-03-26 12:55:24 +08:00
|
|
|
enum object_type type;
|
|
|
|
unsigned long size;
|
2023-03-28 21:58:50 +08:00
|
|
|
char *buf = repo_read_object_file(the_repository,
|
|
|
|
&commit->object.oid, &type,
|
|
|
|
&size);
|
2010-07-22 21:18:33 +08:00
|
|
|
const char *subject_start;
|
|
|
|
int subject_len;
|
2009-03-26 12:55:24 +08:00
|
|
|
|
2024-02-05 22:35:53 +08:00
|
|
|
if (!buf)
|
|
|
|
die(_("unable to read %s"), oid_to_hex(&commit->object.oid));
|
|
|
|
|
2009-03-26 12:55:24 +08:00
|
|
|
fprintf(stderr, "%c%c%c ",
|
2020-08-08 05:58:38 +08:00
|
|
|
(commit_flags & TREESAME) ? ' ' : 'T',
|
|
|
|
(commit_flags & UNINTERESTING) ? 'U' : ' ',
|
|
|
|
(commit_flags & COUNTED) ? 'C' : ' ');
|
2018-09-02 15:42:50 +08:00
|
|
|
if (*commit_weight_at(&commit_weight, p->item))
|
2009-03-26 12:55:24 +08:00
|
|
|
fprintf(stderr, "%3d", weight(p));
|
|
|
|
else
|
|
|
|
fprintf(stderr, "---");
|
2018-03-25 18:57:36 +08:00
|
|
|
fprintf(stderr, " %.*s", 8, oid_to_hex(&commit->object.oid));
|
2009-03-26 12:55:24 +08:00
|
|
|
for (pp = commit->parents; pp; pp = pp->next)
|
|
|
|
fprintf(stderr, " %.*s", 8,
|
2018-03-25 18:57:36 +08:00
|
|
|
oid_to_hex(&pp->item->object.oid));
|
2009-03-26 12:55:24 +08:00
|
|
|
|
2010-07-22 21:18:33 +08:00
|
|
|
subject_len = find_commit_subject(buf, &subject_start);
|
|
|
|
if (subject_len)
|
|
|
|
fprintf(stderr, " %.*s", subject_len, subject_start);
|
2009-03-26 12:55:24 +08:00
|
|
|
fprintf(stderr, "\n");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct commit_list *best_bisection(struct commit_list *list, int nr)
|
|
|
|
{
|
|
|
|
struct commit_list *p, *best;
|
|
|
|
int best_distance = -1;
|
|
|
|
|
|
|
|
best = list;
|
|
|
|
for (p = list; p; p = p->next) {
|
|
|
|
int distance;
|
2020-08-08 05:58:38 +08:00
|
|
|
unsigned commit_flags = p->item->object.flags;
|
2009-03-26 12:55:24 +08:00
|
|
|
|
2020-08-08 05:58:38 +08:00
|
|
|
if (commit_flags & TREESAME)
|
2009-03-26 12:55:24 +08:00
|
|
|
continue;
|
|
|
|
distance = weight(p);
|
|
|
|
if (nr - distance < distance)
|
|
|
|
distance = nr - distance;
|
|
|
|
if (distance > best_distance) {
|
|
|
|
best = p;
|
|
|
|
best_distance = distance;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return best;
|
|
|
|
}
|
|
|
|
|
|
|
|
struct commit_dist {
|
|
|
|
struct commit *commit;
|
|
|
|
int distance;
|
|
|
|
};
|
|
|
|
|
|
|
|
static int compare_commit_dist(const void *a_, const void *b_)
|
|
|
|
{
|
|
|
|
struct commit_dist *a, *b;
|
|
|
|
|
|
|
|
a = (struct commit_dist *)a_;
|
|
|
|
b = (struct commit_dist *)b_;
|
|
|
|
if (a->distance != b->distance)
|
|
|
|
return b->distance - a->distance; /* desc sort */
|
2015-11-10 10:22:28 +08:00
|
|
|
return oidcmp(&a->commit->object.oid, &b->commit->object.oid);
|
2009-03-26 12:55:24 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static struct commit_list *best_bisection_sorted(struct commit_list *list, int nr)
|
|
|
|
{
|
|
|
|
struct commit_list *p;
|
|
|
|
struct commit_dist *array = xcalloc(nr, sizeof(*array));
|
2017-03-29 03:46:50 +08:00
|
|
|
struct strbuf buf = STRBUF_INIT;
|
2009-03-26 12:55:24 +08:00
|
|
|
int cnt, i;
|
|
|
|
|
|
|
|
for (p = list, cnt = 0; p; p = p->next) {
|
|
|
|
int distance;
|
2020-08-08 05:58:38 +08:00
|
|
|
unsigned commit_flags = p->item->object.flags;
|
2009-03-26 12:55:24 +08:00
|
|
|
|
2020-08-08 05:58:38 +08:00
|
|
|
if (commit_flags & TREESAME)
|
2009-03-26 12:55:24 +08:00
|
|
|
continue;
|
|
|
|
distance = weight(p);
|
|
|
|
if (nr - distance < distance)
|
|
|
|
distance = nr - distance;
|
|
|
|
array[cnt].commit = p->item;
|
|
|
|
array[cnt].distance = distance;
|
|
|
|
cnt++;
|
|
|
|
}
|
2016-09-29 23:27:31 +08:00
|
|
|
QSORT(array, cnt, compare_commit_dist);
|
2009-03-26 12:55:24 +08:00
|
|
|
for (p = list, i = 0; i < cnt; i++) {
|
|
|
|
struct object *obj = &(array[i].commit->object);
|
|
|
|
|
2017-03-29 03:46:50 +08:00
|
|
|
strbuf_reset(&buf);
|
|
|
|
strbuf_addf(&buf, "dist=%d", array[i].distance);
|
|
|
|
add_name_decoration(DECORATION_NONE, buf.buf, obj);
|
2014-08-26 18:23:36 +08:00
|
|
|
|
2009-03-26 12:55:24 +08:00
|
|
|
p->item = array[i].commit;
|
bisect: fix off-by-one error in `best_bisection_sorted()`
After we have sorted the `cnt`-many commits that we have selected, we
place them into the commit list. We then set `p->next` to NULL, but as
we do so, `p` is already pointing one beyond item number `cnt`. Indeed,
we check whether `p` is NULL before dereferencing it.
This only matters if there are TREESAME-commits. Since they should be
skipped, they are not included in `cnt` and we will hit the situation
where we set `p->next` to NULL. As a result, the list will be one longer
than it should be. The last commit in the list will be one which occurs
earlier, or which shouldn't be included.
Do not update `p` the very last round in the loop. This ensures that
after the loop, `p->next` points to the remainder of the list, and we
can set it to NULL. While we're here, free that remainder to fix a
memory leak.
Signed-off-by: Martin Ågren <martin.agren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-11-06 04:24:30 +08:00
|
|
|
if (i < cnt - 1)
|
|
|
|
p = p->next;
|
2009-03-26 12:55:24 +08:00
|
|
|
}
|
2018-01-04 02:48:52 +08:00
|
|
|
if (p) {
|
|
|
|
free_commit_list(p->next);
|
|
|
|
p->next = NULL;
|
|
|
|
}
|
2017-03-29 03:46:50 +08:00
|
|
|
strbuf_release(&buf);
|
2009-03-26 12:55:24 +08:00
|
|
|
free(array);
|
|
|
|
return list;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* zero or positive weight is the number of interesting commits it can
|
|
|
|
* reach, including itself. Especially, weight = 0 means it does not
|
|
|
|
* reach any tree-changing commits (e.g. just above uninteresting one
|
|
|
|
* but traversal is with pathspec).
|
|
|
|
*
|
|
|
|
* weight = -1 means it has one parent and its distance is yet to
|
|
|
|
* be computed.
|
|
|
|
*
|
|
|
|
* weight = -2 means it has more than one parent and its distance is
|
|
|
|
* unknown. After running count_distance() first, they will get zero
|
|
|
|
* or positive distance.
|
|
|
|
*/
|
|
|
|
static struct commit_list *do_find_bisection(struct commit_list *list,
|
|
|
|
int nr, int *weights,
|
2020-08-08 05:58:38 +08:00
|
|
|
unsigned bisect_flags)
|
2009-03-26 12:55:24 +08:00
|
|
|
{
|
|
|
|
int n, counted;
|
|
|
|
struct commit_list *p;
|
|
|
|
|
|
|
|
counted = 0;
|
|
|
|
|
|
|
|
for (n = 0, p = list; p; p = p->next) {
|
|
|
|
struct commit *commit = p->item;
|
2020-08-08 05:58:38 +08:00
|
|
|
unsigned commit_flags = commit->object.flags;
|
2009-03-26 12:55:24 +08:00
|
|
|
|
2018-05-19 13:28:25 +08:00
|
|
|
*commit_weight_at(&commit_weight, p->item) = &weights[n++];
|
2020-08-08 05:58:38 +08:00
|
|
|
switch (count_interesting_parents(commit, bisect_flags)) {
|
2009-03-26 12:55:24 +08:00
|
|
|
case 0:
|
2020-08-08 05:58:38 +08:00
|
|
|
if (!(commit_flags & TREESAME)) {
|
2009-03-26 12:55:24 +08:00
|
|
|
weight_set(p, 1);
|
|
|
|
counted++;
|
|
|
|
show_list("bisection 2 count one",
|
|
|
|
counted, nr, list);
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* otherwise, it is known not to reach any
|
|
|
|
* tree-changing commit and gets weight 0.
|
|
|
|
*/
|
|
|
|
break;
|
|
|
|
case 1:
|
|
|
|
weight_set(p, -1);
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
weight_set(p, -2);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
show_list("bisection 2 initialize", counted, nr, list);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If you have only one parent in the resulting set
|
|
|
|
* then you can reach one commit more than that parent
|
|
|
|
* 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
|
|
|
|
* 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.
|
|
|
|
*
|
|
|
|
* So we will first count distance of merges the usual
|
|
|
|
* 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;
|
2020-08-08 05:58:38 +08:00
|
|
|
if (bisect_flags & FIND_BISECTION_FIRST_PARENT_ONLY)
|
2020-08-08 05:58:35 +08:00
|
|
|
BUG("shouldn't be calling count-distance in fp mode");
|
2009-03-26 12:55:24 +08:00
|
|
|
weight_set(p, count_distance(p));
|
|
|
|
clear_distance(list);
|
|
|
|
|
bisect: loosen halfway() check for a large number of commits
'git bisect start ...' and subsequent 'git bisect (good|bad)' commands
can take quite a while when the given/remaining revision range between
good and bad commits is big and contains a lot of merge commits, e.g.
in git.git:
$ git rev-list --count v1.6.0..v2.28.0
44284
$ time git bisect start v2.28.0 v1.6.0
Bisecting: 22141 revisions left to test after this (roughly 15 steps)
[e197c21807dacadc8305250baa0b9228819189d4] unable_to_lock_die(): rename function from unable_to_lock_index_die()
real 0m15.472s
user 0m15.220s
sys 0m0.255s
The majority of the runtime is spent in do_find_bisection(), where we
try to find a commit as close as possible to the halfway point between
the bad and good revisions, i.e. a commit from which the number of
reachable commits that are in the good-bad range is half the total
number of commits in that range. So we count how many commits are
reachable in the good-bad range for each commit in that range, which
is quick and easy for a linear history, even over 300k commits in a
linear range are handled in ~0.3s on my machine. Alas, handling merge
commits is non-trivial and quite expensive as the algorithm used seems
to be quadratic, causing the long runtime shown above.
Interestingly, look at what a big difference one additional commit
can make:
$ git rev-list --count v1.6.0^..v2.28.0
44285
$ time git bisect start v2.28.0 v1.6.0^
Bisecting: 22142 revisions left to test after this (roughly 15 steps)
[565301e41670825ceedf75220f2918ae76831240] Sync with 2.1.2
real 0m5.848s
user 0m5.600s
sys 0m0.252s
The difference is caused by one of the optimizations attempting to cut
down the runtime added in 1c4fea3a40 (git-rev-list --bisect:
optimization, 2007-03-21):
Another small optimization is whenever we find a half-way commit
(that is, a commit that can reach exactly half of the commits),
we stop giving counts to remaining commits, as we will not find
any better commit than we just found.
In this second 'git bisect start' command we happen to find a commit
exactly at the halfway point and can return early, but in the first
case there is no such commit, so we can't return early and end up
counting the number of reachable commits from all commits in the
good-bad range.
However, when we have thousands of commits it's not all that important
to find the _exact_ halfway point, a few commits more or less doesn't
make any real difference for the bisection.
So let's loosen the check in the halfway() helper to consider commits
within about 0.1% of the exact halfway point as halfway as well, and
rename the function to approx_halfway() accordingly. This will allow
us to return early on a bigger good-bad range, even when there is no
commit exactly at the halfway point, thereby reducing the runtime of
the first command above considerably, from ~15s to 4.901s.
Furthermore, even if there is a commit exactly at the halfway point,
we might still stumble upon a commit within that 0.1% range before
finding the exact halfway point, allowing us to return a bit earlier,
slightly reducing the runtime of the second command from 5.848s to
5.058s. Note that this change doesn't affect good-bad ranges
containing ~2000 commits or less, because that 0.1% tolerance becomes
zero due to integer arithmetic; however, if the range is that small
then counting the reachable commits for all commits is already fast
enough anyway.
Naturally, this will likely change which commits get picked at each
bisection step, and, in turn, might change how many bisection steps
are necessary to find the first bad commit. If the number of
necessary bisection steps were to increase often, then this change
could backfire, because building and testing at each step might take
much longer than the time spared. OTOH, if the number of steps were
to decrease, then it would be a double win.
So I ran some tests to see how often that happens: picked random good
and bad starting revisions at least 50k commits apart and a random
first bad commit in between in git.git, and used 'git bisect run git
merge-base --is-ancestor HEAD $first_bad_commit' to check the number
of necessary bisection steps. After repeating all this 1000 times
both with and without this patch I found that:
- 146 cases needed one more bisection step than before, 149 cases
needed one less step, while in the remaining 705 cases the number
of steps didn't change. So the number of bisection steps does
indeed change in a non-negligible number of cases, but it seems
that the average number of steps doesn't change in the long run.
- The first 'git bisect start' command got over 3x faster in 456
cases, so this "no commit at the exact halfway point" case seems
to be common enough to care about.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-11-13 00:19:38 +08:00
|
|
|
/* Does it happen to be at half-way? */
|
|
|
|
if (!(bisect_flags & FIND_BISECTION_ALL) &&
|
|
|
|
approx_halfway(p, nr))
|
2009-03-26 12:55:24 +08:00
|
|
|
return p;
|
|
|
|
counted++;
|
|
|
|
}
|
|
|
|
|
|
|
|
show_list("bisection 2 count_distance", counted, nr, list);
|
|
|
|
|
|
|
|
while (counted < nr) {
|
|
|
|
for (p = list; p; p = p->next) {
|
|
|
|
struct commit_list *q;
|
2020-08-08 05:58:38 +08:00
|
|
|
unsigned commit_flags = p->item->object.flags;
|
2009-03-26 12:55:24 +08:00
|
|
|
|
|
|
|
if (0 <= weight(p))
|
|
|
|
continue;
|
2020-08-08 05:58:35 +08:00
|
|
|
|
|
|
|
for (q = p->item->parents;
|
|
|
|
q;
|
2020-08-08 05:58:38 +08:00
|
|
|
q = bisect_flags & FIND_BISECTION_FIRST_PARENT_ONLY ? NULL : q->next) {
|
2009-03-26 12:55:24 +08:00
|
|
|
if (q->item->object.flags & UNINTERESTING)
|
|
|
|
continue;
|
|
|
|
if (0 <= weight(q))
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (!q)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* weight for p is unknown but q is known.
|
|
|
|
* add one for p itself if p is to be counted,
|
|
|
|
* otherwise inherit it from q directly.
|
|
|
|
*/
|
2020-08-08 05:58:38 +08:00
|
|
|
if (!(commit_flags & TREESAME)) {
|
2009-03-26 12:55:24 +08:00
|
|
|
weight_set(p, weight(q)+1);
|
|
|
|
counted++;
|
|
|
|
show_list("bisection 2 count one",
|
|
|
|
counted, nr, list);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
weight_set(p, weight(q));
|
|
|
|
|
bisect: loosen halfway() check for a large number of commits
'git bisect start ...' and subsequent 'git bisect (good|bad)' commands
can take quite a while when the given/remaining revision range between
good and bad commits is big and contains a lot of merge commits, e.g.
in git.git:
$ git rev-list --count v1.6.0..v2.28.0
44284
$ time git bisect start v2.28.0 v1.6.0
Bisecting: 22141 revisions left to test after this (roughly 15 steps)
[e197c21807dacadc8305250baa0b9228819189d4] unable_to_lock_die(): rename function from unable_to_lock_index_die()
real 0m15.472s
user 0m15.220s
sys 0m0.255s
The majority of the runtime is spent in do_find_bisection(), where we
try to find a commit as close as possible to the halfway point between
the bad and good revisions, i.e. a commit from which the number of
reachable commits that are in the good-bad range is half the total
number of commits in that range. So we count how many commits are
reachable in the good-bad range for each commit in that range, which
is quick and easy for a linear history, even over 300k commits in a
linear range are handled in ~0.3s on my machine. Alas, handling merge
commits is non-trivial and quite expensive as the algorithm used seems
to be quadratic, causing the long runtime shown above.
Interestingly, look at what a big difference one additional commit
can make:
$ git rev-list --count v1.6.0^..v2.28.0
44285
$ time git bisect start v2.28.0 v1.6.0^
Bisecting: 22142 revisions left to test after this (roughly 15 steps)
[565301e41670825ceedf75220f2918ae76831240] Sync with 2.1.2
real 0m5.848s
user 0m5.600s
sys 0m0.252s
The difference is caused by one of the optimizations attempting to cut
down the runtime added in 1c4fea3a40 (git-rev-list --bisect:
optimization, 2007-03-21):
Another small optimization is whenever we find a half-way commit
(that is, a commit that can reach exactly half of the commits),
we stop giving counts to remaining commits, as we will not find
any better commit than we just found.
In this second 'git bisect start' command we happen to find a commit
exactly at the halfway point and can return early, but in the first
case there is no such commit, so we can't return early and end up
counting the number of reachable commits from all commits in the
good-bad range.
However, when we have thousands of commits it's not all that important
to find the _exact_ halfway point, a few commits more or less doesn't
make any real difference for the bisection.
So let's loosen the check in the halfway() helper to consider commits
within about 0.1% of the exact halfway point as halfway as well, and
rename the function to approx_halfway() accordingly. This will allow
us to return early on a bigger good-bad range, even when there is no
commit exactly at the halfway point, thereby reducing the runtime of
the first command above considerably, from ~15s to 4.901s.
Furthermore, even if there is a commit exactly at the halfway point,
we might still stumble upon a commit within that 0.1% range before
finding the exact halfway point, allowing us to return a bit earlier,
slightly reducing the runtime of the second command from 5.848s to
5.058s. Note that this change doesn't affect good-bad ranges
containing ~2000 commits or less, because that 0.1% tolerance becomes
zero due to integer arithmetic; however, if the range is that small
then counting the reachable commits for all commits is already fast
enough anyway.
Naturally, this will likely change which commits get picked at each
bisection step, and, in turn, might change how many bisection steps
are necessary to find the first bad commit. If the number of
necessary bisection steps were to increase often, then this change
could backfire, because building and testing at each step might take
much longer than the time spared. OTOH, if the number of steps were
to decrease, then it would be a double win.
So I ran some tests to see how often that happens: picked random good
and bad starting revisions at least 50k commits apart and a random
first bad commit in between in git.git, and used 'git bisect run git
merge-base --is-ancestor HEAD $first_bad_commit' to check the number
of necessary bisection steps. After repeating all this 1000 times
both with and without this patch I found that:
- 146 cases needed one more bisection step than before, 149 cases
needed one less step, while in the remaining 705 cases the number
of steps didn't change. So the number of bisection steps does
indeed change in a non-negligible number of cases, but it seems
that the average number of steps doesn't change in the long run.
- The first 'git bisect start' command got over 3x faster in 456
cases, so this "no commit at the exact halfway point" case seems
to be common enough to care about.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-11-13 00:19:38 +08:00
|
|
|
/* Does it happen to be at half-way? */
|
|
|
|
if (!(bisect_flags & FIND_BISECTION_ALL) &&
|
|
|
|
approx_halfway(p, nr))
|
2009-03-26 12:55:24 +08:00
|
|
|
return p;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
show_list("bisection 2 counted all", counted, nr, list);
|
|
|
|
|
2020-08-08 05:58:38 +08:00
|
|
|
if (!(bisect_flags & FIND_BISECTION_ALL))
|
2009-03-26 12:55:24 +08:00
|
|
|
return best_bisection(list, nr);
|
|
|
|
else
|
|
|
|
return best_bisection_sorted(list, nr);
|
|
|
|
}
|
|
|
|
|
2017-11-06 04:24:28 +08:00
|
|
|
void find_bisection(struct commit_list **commit_list, int *reaches,
|
2020-08-08 05:58:38 +08:00
|
|
|
int *all, unsigned bisect_flags)
|
2009-03-26 12:55:24 +08:00
|
|
|
{
|
|
|
|
int nr, on_list;
|
2017-11-06 04:24:28 +08:00
|
|
|
struct commit_list *list, *p, *best, *next, *last;
|
2009-03-26 12:55:24 +08:00
|
|
|
int *weights;
|
|
|
|
|
2017-11-06 04:24:28 +08:00
|
|
|
show_list("bisection 2 entry", 0, 0, *commit_list);
|
2018-05-19 13:28:25 +08:00
|
|
|
init_commit_weight(&commit_weight);
|
2009-03-26 12:55:24 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Count the number of total and tree-changing items on the
|
|
|
|
* list, while reversing the list.
|
|
|
|
*/
|
2017-11-06 04:24:28 +08:00
|
|
|
for (nr = on_list = 0, last = NULL, p = *commit_list;
|
2009-03-26 12:55:24 +08:00
|
|
|
p;
|
|
|
|
p = next) {
|
2020-08-08 05:58:38 +08:00
|
|
|
unsigned commit_flags = p->item->object.flags;
|
2009-03-26 12:55:24 +08:00
|
|
|
|
|
|
|
next = p->next;
|
2020-08-08 05:58:38 +08:00
|
|
|
if (commit_flags & UNINTERESTING) {
|
2017-11-06 04:24:29 +08:00
|
|
|
free(p);
|
2009-03-26 12:55:24 +08:00
|
|
|
continue;
|
2017-11-06 04:24:29 +08:00
|
|
|
}
|
2009-03-26 12:55:24 +08:00
|
|
|
p->next = last;
|
|
|
|
last = p;
|
2020-08-08 05:58:38 +08:00
|
|
|
if (!(commit_flags & TREESAME))
|
2009-03-26 12:55:24 +08:00
|
|
|
nr++;
|
|
|
|
on_list++;
|
|
|
|
}
|
|
|
|
list = last;
|
|
|
|
show_list("bisection 2 sorted", 0, nr, list);
|
|
|
|
|
|
|
|
*all = nr;
|
2021-03-14 00:17:22 +08:00
|
|
|
CALLOC_ARRAY(weights, on_list);
|
2009-03-26 12:55:24 +08:00
|
|
|
|
|
|
|
/* Do the real work of finding bisection commit. */
|
2020-08-08 05:58:38 +08:00
|
|
|
best = do_find_bisection(list, nr, weights, bisect_flags);
|
2009-03-26 12:55:24 +08:00
|
|
|
if (best) {
|
2020-08-08 05:58:38 +08:00
|
|
|
if (!(bisect_flags & FIND_BISECTION_ALL)) {
|
2017-11-06 04:24:31 +08:00
|
|
|
list->item = best->item;
|
|
|
|
free_commit_list(list->next);
|
|
|
|
best = list;
|
2009-03-26 12:55:24 +08:00
|
|
|
best->next = NULL;
|
2017-11-06 04:24:31 +08:00
|
|
|
}
|
2009-03-26 12:55:24 +08:00
|
|
|
*reaches = weight(best);
|
|
|
|
}
|
|
|
|
free(weights);
|
2017-11-06 04:24:28 +08:00
|
|
|
*commit_list = best;
|
2018-05-19 13:28:25 +08:00
|
|
|
clear_commit_weight(&commit_weight);
|
2009-03-26 12:55:24 +08:00
|
|
|
}
|
|
|
|
|
2024-08-09 23:37:50 +08:00
|
|
|
static int register_ref(const char *refname, const char *referent UNUSED, const struct object_id *oid,
|
2022-08-26 01:09:48 +08:00
|
|
|
int flags UNUSED, void *cb_data UNUSED)
|
2009-03-26 12:55:54 +08:00
|
|
|
{
|
2015-06-29 23:40:29 +08:00
|
|
|
struct strbuf good_prefix = STRBUF_INIT;
|
|
|
|
strbuf_addstr(&good_prefix, term_good);
|
|
|
|
strbuf_addstr(&good_prefix, "-");
|
|
|
|
|
|
|
|
if (!strcmp(refname, term_bad)) {
|
2015-03-14 07:39:29 +08:00
|
|
|
current_bad_oid = xmalloc(sizeof(*current_bad_oid));
|
2015-05-26 02:38:31 +08:00
|
|
|
oidcpy(current_bad_oid, oid);
|
2015-06-29 23:40:29 +08:00
|
|
|
} else if (starts_with(refname, good_prefix.buf)) {
|
2017-03-31 09:40:00 +08:00
|
|
|
oid_array_append(&good_revs, oid);
|
2013-12-01 04:55:40 +08:00
|
|
|
} else if (starts_with(refname, "skip-")) {
|
2017-03-31 09:40:00 +08:00
|
|
|
oid_array_append(&skipped_revs, oid);
|
2009-03-26 12:55:54 +08:00
|
|
|
}
|
|
|
|
|
2015-06-29 23:40:29 +08:00
|
|
|
strbuf_release(&good_prefix);
|
|
|
|
|
2009-03-26 12:55:54 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int read_bisect_refs(void)
|
|
|
|
{
|
2024-05-07 15:11:53 +08:00
|
|
|
return refs_for_each_ref_in(get_main_ref_store(the_repository),
|
|
|
|
"refs/bisect/", register_ref, NULL);
|
2009-03-26 12:55:54 +08:00
|
|
|
}
|
|
|
|
|
memoize common git-path "constant" files
One of the most common uses of git_path() is to pass a
constant, like git_path("MERGE_MSG"). This has two
drawbacks:
1. The return value is a static buffer, and the lifetime
is dependent on other calls to git_path, etc.
2. There's no compile-time checking of the pathname. This
is OK for a one-off (after all, we have to spell it
correctly at least once), but many of these constant
strings appear throughout the code.
This patch introduces a series of functions to "memoize"
these strings, which are essentially globals for the
lifetime of the program. We compute the value once, take
ownership of the buffer, and return the cached value for
subsequent calls. cache.h provides a helper macro for
defining these functions as one-liners, and defines a few
common ones for global use.
Using a macro is a little bit gross, but it does nicely
document the purpose of the functions. If we need to touch
them all later (e.g., because we learned how to change the
git_dir variable at runtime, and need to invalidate all of
the stored values), it will be much easier to have the
complete list.
Note that the shared-global functions have separate, manual
declarations. We could do something clever with the macros
(e.g., expand it to a declaration in some places, and a
declaration _and_ a definition in path.c). But there aren't
that many, and it's probably better to stay away from
too-magical macros.
Likewise, if we abandon the C preprocessor in favor of
generating these with a script, we could get much fancier.
E.g., normalizing "FOO/BAR-BAZ" into "git_path_foo_bar_baz".
But the small amount of saved typing is probably not worth
the resulting confusion to readers who want to grep for the
function's definition.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-08-10 17:38:57 +08:00
|
|
|
static GIT_PATH_FUNC(git_path_bisect_names, "BISECT_NAMES")
|
2017-09-29 14:49:39 +08:00
|
|
|
static GIT_PATH_FUNC(git_path_bisect_ancestors_ok, "BISECT_ANCESTORS_OK")
|
|
|
|
static GIT_PATH_FUNC(git_path_bisect_run, "BISECT_RUN")
|
|
|
|
static GIT_PATH_FUNC(git_path_bisect_start, "BISECT_START")
|
|
|
|
static GIT_PATH_FUNC(git_path_bisect_log, "BISECT_LOG")
|
2017-04-21 05:08:25 +08:00
|
|
|
static GIT_PATH_FUNC(git_path_bisect_terms, "BISECT_TERMS")
|
2020-08-08 05:58:37 +08:00
|
|
|
static GIT_PATH_FUNC(git_path_bisect_first_parent, "BISECT_FIRST_PARENT")
|
memoize common git-path "constant" files
One of the most common uses of git_path() is to pass a
constant, like git_path("MERGE_MSG"). This has two
drawbacks:
1. The return value is a static buffer, and the lifetime
is dependent on other calls to git_path, etc.
2. There's no compile-time checking of the pathname. This
is OK for a one-off (after all, we have to spell it
correctly at least once), but many of these constant
strings appear throughout the code.
This patch introduces a series of functions to "memoize"
these strings, which are essentially globals for the
lifetime of the program. We compute the value once, take
ownership of the buffer, and return the cached value for
subsequent calls. cache.h provides a helper macro for
defining these functions as one-liners, and defines a few
common ones for global use.
Using a macro is a little bit gross, but it does nicely
document the purpose of the functions. If we need to touch
them all later (e.g., because we learned how to change the
git_dir variable at runtime, and need to invalidate all of
the stored values), it will be much easier to have the
complete list.
Note that the shared-global functions have separate, manual
declarations. We could do something clever with the macros
(e.g., expand it to a declaration in some places, and a
declaration _and_ a definition in path.c). But there aren't
that many, and it's probably better to stay away from
too-magical macros.
Likewise, if we abandon the C preprocessor in favor of
generating these with a script, we could get much fancier.
E.g., normalizing "FOO/BAR-BAZ" into "git_path_foo_bar_baz".
But the small amount of saved typing is probably not worth
the resulting confusion to readers who want to grep for the
function's definition.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-08-10 17:38:57 +08:00
|
|
|
|
2020-07-29 04:24:53 +08:00
|
|
|
static void read_bisect_paths(struct strvec *array)
|
2009-03-26 12:55:59 +08:00
|
|
|
{
|
|
|
|
struct strbuf str = STRBUF_INIT;
|
memoize common git-path "constant" files
One of the most common uses of git_path() is to pass a
constant, like git_path("MERGE_MSG"). This has two
drawbacks:
1. The return value is a static buffer, and the lifetime
is dependent on other calls to git_path, etc.
2. There's no compile-time checking of the pathname. This
is OK for a one-off (after all, we have to spell it
correctly at least once), but many of these constant
strings appear throughout the code.
This patch introduces a series of functions to "memoize"
these strings, which are essentially globals for the
lifetime of the program. We compute the value once, take
ownership of the buffer, and return the cached value for
subsequent calls. cache.h provides a helper macro for
defining these functions as one-liners, and defines a few
common ones for global use.
Using a macro is a little bit gross, but it does nicely
document the purpose of the functions. If we need to touch
them all later (e.g., because we learned how to change the
git_dir variable at runtime, and need to invalidate all of
the stored values), it will be much easier to have the
complete list.
Note that the shared-global functions have separate, manual
declarations. We could do something clever with the macros
(e.g., expand it to a declaration in some places, and a
declaration _and_ a definition in path.c). But there aren't
that many, and it's probably better to stay away from
too-magical macros.
Likewise, if we abandon the C preprocessor in favor of
generating these with a script, we could get much fancier.
E.g., normalizing "FOO/BAR-BAZ" into "git_path_foo_bar_baz".
But the small amount of saved typing is probably not worth
the resulting confusion to readers who want to grep for the
function's definition.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-08-10 17:38:57 +08:00
|
|
|
const char *filename = git_path_bisect_names();
|
2017-05-03 18:16:46 +08:00
|
|
|
FILE *fp = xfopen(filename, "r");
|
2009-03-26 12:55:59 +08:00
|
|
|
|
2016-01-14 07:31:17 +08:00
|
|
|
while (strbuf_getline_lf(&str, fp) != EOF) {
|
2009-03-26 12:55:59 +08:00
|
|
|
strbuf_trim(&str);
|
2020-07-29 04:24:02 +08:00
|
|
|
if (sq_dequote_to_strvec(str.buf, array))
|
2016-06-18 04:21:12 +08:00
|
|
|
die(_("Badly quoted content in file '%s': %s"),
|
2011-09-14 05:58:14 +08:00
|
|
|
filename, str.buf);
|
2009-03-26 12:55:59 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
strbuf_release(&str);
|
|
|
|
fclose(fp);
|
|
|
|
}
|
|
|
|
|
2020-03-30 22:04:06 +08:00
|
|
|
static char *join_oid_array_hex(struct oid_array *array, char delim)
|
2009-05-09 23:55:45 +08:00
|
|
|
{
|
|
|
|
struct strbuf joined_hexs = STRBUF_INIT;
|
|
|
|
int i;
|
|
|
|
|
2011-05-20 05:34:33 +08:00
|
|
|
for (i = 0; i < array->nr; i++) {
|
2017-03-27 00:01:37 +08:00
|
|
|
strbuf_addstr(&joined_hexs, oid_to_hex(array->oid + i));
|
2011-05-20 05:34:33 +08:00
|
|
|
if (i + 1 < array->nr)
|
2009-05-09 23:55:45 +08:00
|
|
|
strbuf_addch(&joined_hexs, delim);
|
|
|
|
}
|
|
|
|
|
|
|
|
return strbuf_detach(&joined_hexs, NULL);
|
|
|
|
}
|
|
|
|
|
2009-06-06 12:41:33 +08:00
|
|
|
/*
|
|
|
|
* In this function, passing a not NULL skipped_first is very special.
|
|
|
|
* It means that we want to know if the first commit in the list is
|
|
|
|
* skipped because we will want to test a commit away from it if it is
|
|
|
|
* indeed skipped.
|
|
|
|
* So if the first commit is skipped, we cannot take the shortcut to
|
|
|
|
* just "return list" when we find the first non skipped commit, we
|
|
|
|
* have to return a fully filtered list.
|
|
|
|
*
|
|
|
|
* We use (*skipped_first == -1) to mean "it has been found that the
|
|
|
|
* first commit is not skipped". In this case *skipped_first is set back
|
|
|
|
* to 0 just before the function returns.
|
|
|
|
*/
|
2009-03-26 12:55:49 +08:00
|
|
|
struct commit_list *filter_skipped(struct commit_list *list,
|
|
|
|
struct commit_list **tried,
|
2009-06-06 12:41:33 +08:00
|
|
|
int show_all,
|
|
|
|
int *count,
|
|
|
|
int *skipped_first)
|
2009-03-26 12:55:49 +08:00
|
|
|
{
|
|
|
|
struct commit_list *filtered = NULL, **f = &filtered;
|
|
|
|
|
|
|
|
*tried = NULL;
|
|
|
|
|
2009-06-06 12:41:33 +08:00
|
|
|
if (skipped_first)
|
|
|
|
*skipped_first = 0;
|
|
|
|
if (count)
|
|
|
|
*count = 0;
|
|
|
|
|
2011-05-20 05:34:33 +08:00
|
|
|
if (!skipped_revs.nr)
|
2009-03-26 12:55:49 +08:00
|
|
|
return list;
|
|
|
|
|
|
|
|
while (list) {
|
|
|
|
struct commit_list *next = list->next;
|
|
|
|
list->next = NULL;
|
2017-03-31 09:40:00 +08:00
|
|
|
if (0 <= oid_array_lookup(&skipped_revs, &list->item->object.oid)) {
|
2009-06-06 12:41:33 +08:00
|
|
|
if (skipped_first && !*skipped_first)
|
|
|
|
*skipped_first = 1;
|
2009-03-26 12:55:49 +08:00
|
|
|
/* Move current to tried list */
|
|
|
|
*tried = list;
|
|
|
|
tried = &list->next;
|
|
|
|
} else {
|
2009-06-06 12:41:33 +08:00
|
|
|
if (!show_all) {
|
|
|
|
if (!skipped_first || !*skipped_first)
|
|
|
|
return list;
|
|
|
|
} else if (skipped_first && !*skipped_first) {
|
|
|
|
/* This means we know it's not skipped */
|
|
|
|
*skipped_first = -1;
|
|
|
|
}
|
2009-03-26 12:55:49 +08:00
|
|
|
/* Move current to filtered list */
|
|
|
|
*f = list;
|
|
|
|
f = &list->next;
|
2009-06-06 12:41:33 +08:00
|
|
|
if (count)
|
|
|
|
(*count)++;
|
2009-03-26 12:55:49 +08:00
|
|
|
}
|
|
|
|
list = next;
|
|
|
|
}
|
|
|
|
|
2009-06-06 12:41:33 +08:00
|
|
|
if (skipped_first && *skipped_first == -1)
|
|
|
|
*skipped_first = 0;
|
|
|
|
|
2009-03-26 12:55:49 +08:00
|
|
|
return filtered;
|
|
|
|
}
|
2009-03-26 12:55:54 +08:00
|
|
|
|
2009-06-13 13:21:06 +08:00
|
|
|
#define PRN_MODULO 32768
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This is a pseudo random number generator based on "man 3 rand".
|
|
|
|
* It is not used properly because the seed is the argument and it
|
|
|
|
* is increased by one between each call, but that should not matter
|
|
|
|
* for this application.
|
|
|
|
*/
|
2018-12-09 18:25:21 +08:00
|
|
|
static unsigned get_prn(unsigned count)
|
|
|
|
{
|
2009-06-13 13:21:06 +08:00
|
|
|
count = count * 1103515245 + 12345;
|
2013-04-04 03:17:55 +08:00
|
|
|
return (count/65536) % PRN_MODULO;
|
2009-06-13 13:21:06 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Custom integer square root from
|
2017-05-13 17:54:51 +08:00
|
|
|
* https://en.wikipedia.org/wiki/Integer_square_root
|
2009-06-13 13:21:06 +08:00
|
|
|
*/
|
|
|
|
static int sqrti(int val)
|
|
|
|
{
|
|
|
|
float d, x = val;
|
|
|
|
|
2020-02-17 16:40:30 +08:00
|
|
|
if (!val)
|
2009-06-13 13:21:06 +08:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
do {
|
|
|
|
float y = (x + (float)val / x) / 2;
|
|
|
|
d = (y > x) ? y - x : x - y;
|
|
|
|
x = y;
|
|
|
|
} while (d >= 0.5);
|
|
|
|
|
|
|
|
return (int)x;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct commit_list *skip_away(struct commit_list *list, int count)
|
2009-06-06 12:41:34 +08:00
|
|
|
{
|
|
|
|
struct commit_list *cur, *previous;
|
2009-06-13 13:21:06 +08:00
|
|
|
int prn, index, i;
|
|
|
|
|
|
|
|
prn = get_prn(count);
|
|
|
|
index = (count * prn / PRN_MODULO) * sqrti(prn) / sqrti(PRN_MODULO);
|
2009-06-06 12:41:34 +08:00
|
|
|
|
|
|
|
cur = list;
|
|
|
|
previous = NULL;
|
|
|
|
|
|
|
|
for (i = 0; cur; cur = cur->next, i++) {
|
|
|
|
if (i == index) {
|
2018-08-29 05:22:48 +08:00
|
|
|
if (!oideq(&cur->item->object.oid, current_bad_oid))
|
2009-06-06 12:41:34 +08:00
|
|
|
return cur;
|
|
|
|
if (previous)
|
|
|
|
return previous;
|
|
|
|
return list;
|
|
|
|
}
|
|
|
|
previous = cur;
|
|
|
|
}
|
|
|
|
|
|
|
|
return list;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct commit_list *managed_skipped(struct commit_list *list,
|
|
|
|
struct commit_list **tried)
|
|
|
|
{
|
|
|
|
int count, skipped_first;
|
|
|
|
|
|
|
|
*tried = NULL;
|
|
|
|
|
2011-05-20 05:34:33 +08:00
|
|
|
if (!skipped_revs.nr)
|
2009-06-06 12:41:34 +08:00
|
|
|
return list;
|
|
|
|
|
|
|
|
list = filter_skipped(list, tried, 0, &count, &skipped_first);
|
|
|
|
|
|
|
|
if (!skipped_first)
|
|
|
|
return list;
|
|
|
|
|
2009-06-13 13:21:06 +08:00
|
|
|
return skip_away(list, count);
|
2009-06-06 12:41:34 +08:00
|
|
|
}
|
|
|
|
|
2018-11-10 13:48:59 +08:00
|
|
|
static void bisect_rev_setup(struct repository *r, struct rev_info *revs,
|
2022-08-02 23:33:15 +08:00
|
|
|
struct strvec *rev_argv,
|
2018-11-10 13:48:59 +08:00
|
|
|
const char *prefix,
|
2009-05-17 23:36:44 +08:00
|
|
|
const char *bad_format, const char *good_format,
|
|
|
|
int read_paths)
|
2009-03-26 12:55:54 +08:00
|
|
|
{
|
2022-08-02 23:33:16 +08:00
|
|
|
struct setup_revision_opt opt = {
|
|
|
|
.free_removed_argv_elements = 1,
|
|
|
|
};
|
2009-05-09 23:55:40 +08:00
|
|
|
int i;
|
|
|
|
|
2018-11-10 13:48:59 +08:00
|
|
|
repo_init_revisions(r, revs, prefix);
|
2009-03-26 12:55:59 +08:00
|
|
|
revs->abbrev = 0;
|
|
|
|
revs->commit_format = CMIT_FMT_UNSPECIFIED;
|
2009-03-26 12:55:54 +08:00
|
|
|
|
2009-05-09 23:55:41 +08:00
|
|
|
/* rev_argv.argv[0] will be ignored by setup_revisions */
|
2022-08-02 23:33:15 +08:00
|
|
|
strvec_push(rev_argv, "bisect_rev_setup");
|
|
|
|
strvec_pushf(rev_argv, bad_format, oid_to_hex(current_bad_oid));
|
2011-05-20 05:34:33 +08:00
|
|
|
for (i = 0; i < good_revs.nr; i++)
|
2022-08-02 23:33:15 +08:00
|
|
|
strvec_pushf(rev_argv, good_format,
|
strvec: fix indentation in renamed calls
Code which split an argv_array call across multiple lines, like:
argv_array_pushl(&args, "one argument",
"another argument", "and more",
NULL);
was recently mechanically renamed to use strvec, which results in
mis-matched indentation like:
strvec_pushl(&args, "one argument",
"another argument", "and more",
NULL);
Let's fix these up to align the arguments with the opening paren. I did
this manually by sifting through the results of:
git jump grep 'strvec_.*,$'
and liberally applying my editor's auto-format. Most of the changes are
of the form shown above, though I also normalized a few that had
originally used a single-tab indentation (rather than our usual style of
aligning with the open paren). I also rewrapped a couple of obvious
cases (e.g., where previously too-long lines became short enough to fit
on one), but I wasn't aggressive about it. In cases broken to three or
more lines, the grouping of arguments is sometimes meaningful, and it
wasn't worth my time or reviewer time to ponder each case individually.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-07-29 04:26:31 +08:00
|
|
|
oid_to_hex(good_revs.oid + i));
|
2022-08-02 23:33:15 +08:00
|
|
|
strvec_push(rev_argv, "--");
|
2009-05-17 23:36:44 +08:00
|
|
|
if (read_paths)
|
2022-08-02 23:33:15 +08:00
|
|
|
read_bisect_paths(rev_argv);
|
2009-03-26 12:55:59 +08:00
|
|
|
|
2022-08-02 23:33:16 +08:00
|
|
|
setup_revisions(rev_argv->nr, rev_argv->v, revs, &opt);
|
2009-03-26 12:55:59 +08:00
|
|
|
}
|
|
|
|
|
2009-05-17 23:36:44 +08:00
|
|
|
static void bisect_common(struct rev_info *revs)
|
2009-04-19 17:55:57 +08:00
|
|
|
{
|
|
|
|
if (prepare_revision_walk(revs))
|
|
|
|
die("revision walk setup failed");
|
|
|
|
if (revs->tree_objects)
|
2019-01-17 02:25:58 +08:00
|
|
|
mark_edges_uninteresting(revs, NULL, 0);
|
2009-04-19 17:55:57 +08:00
|
|
|
}
|
|
|
|
|
2020-02-17 16:40:34 +08:00
|
|
|
static enum bisect_error error_if_skipped_commits(struct commit_list *tried,
|
2015-03-14 07:39:29 +08:00
|
|
|
const struct object_id *bad)
|
2009-04-19 17:56:07 +08:00
|
|
|
{
|
|
|
|
if (!tried)
|
2020-02-17 16:40:34 +08:00
|
|
|
return BISECT_OK;
|
2009-04-19 17:56:07 +08:00
|
|
|
|
|
|
|
printf("There are only 'skip'ped commits left to test.\n"
|
2015-06-29 23:40:29 +08:00
|
|
|
"The first %s commit could be any of:\n", term_bad);
|
2016-07-09 01:09:28 +08:00
|
|
|
|
|
|
|
for ( ; tried; tried = tried->next)
|
|
|
|
printf("%s\n", oid_to_hex(&tried->item->object.oid));
|
|
|
|
|
2009-04-19 17:56:07 +08:00
|
|
|
if (bad)
|
2015-03-14 07:39:29 +08:00
|
|
|
printf("%s\n", oid_to_hex(bad));
|
2016-06-18 04:21:12 +08:00
|
|
|
printf(_("We cannot bisect more!\n"));
|
2020-02-17 16:40:34 +08:00
|
|
|
|
|
|
|
return BISECT_ONLY_SKIPPED_LEFT;
|
2009-04-19 17:56:07 +08:00
|
|
|
}
|
|
|
|
|
2015-03-14 07:39:29 +08:00
|
|
|
static int is_expected_rev(const struct object_id *oid)
|
2009-05-09 23:55:45 +08:00
|
|
|
{
|
2023-12-14 21:37:13 +08:00
|
|
|
struct object_id expected_oid;
|
2024-05-07 15:11:53 +08:00
|
|
|
if (refs_read_ref(get_main_ref_store(the_repository), "BISECT_EXPECTED_REV", &expected_oid))
|
2009-05-09 23:55:45 +08:00
|
|
|
return 0;
|
2023-12-14 21:37:13 +08:00
|
|
|
return oideq(oid, &expected_oid);
|
2009-05-09 23:55:45 +08:00
|
|
|
}
|
|
|
|
|
bisect--helper: double-check run command on exit code 126 and 127
When a run command cannot be executed or found, shells return exit code
126 or 127, respectively. Valid run commands are allowed to return
these codes as well to indicate bad revisions, though, for historical
reasons. This means typos can cause bogus bisect runs that go over the
full distance and end up reporting invalid results.
The best solution would be to reserve exit codes 126 and 127, like
71b0251cdd (Bisect run: "skip" current commit if script exit code is
125., 2007-10-26) did for 125, and abort bisect run when we get them.
That might be inconvenient for those who relied on the documentation
stating that 126 and 127 can be used for bad revisions, though.
The workaround used by this patch is to run the command on a known-good
revision and abort if we still get the same error code. This adds one
step to runs with scripts that use exit codes 126 and 127, but keeps
them supported, with one exception: It won't work with commands that
cannot recognize the (manually marked) known-good revision as such.
Run commands that use low exit codes are unaffected. Typos are reported
after executing the missing command twice and three checkouts (the first
step, the known good revision and back to the revision of the first
step).
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-01-18 20:46:32 +08:00
|
|
|
enum bisect_error bisect_checkout(const struct object_id *bisect_rev,
|
|
|
|
int no_checkout)
|
2009-04-19 17:56:07 +08:00
|
|
|
{
|
2021-07-28 02:22:18 +08:00
|
|
|
struct commit *commit;
|
|
|
|
struct pretty_print_context pp = {0};
|
|
|
|
struct strbuf commit_msg = STRBUF_INIT;
|
2009-04-19 17:56:07 +08:00
|
|
|
|
2024-05-07 15:11:53 +08:00
|
|
|
refs_update_ref(get_main_ref_store(the_repository), NULL,
|
|
|
|
"BISECT_EXPECTED_REV", bisect_rev, NULL, 0,
|
|
|
|
UPDATE_REFS_DIE_ON_ERR);
|
2009-04-19 17:56:07 +08:00
|
|
|
|
2011-08-04 20:01:00 +08:00
|
|
|
if (no_checkout) {
|
2024-05-07 15:11:53 +08:00
|
|
|
refs_update_ref(get_main_ref_store(the_repository), NULL,
|
|
|
|
"BISECT_HEAD", bisect_rev, NULL, 0,
|
|
|
|
UPDATE_REFS_DIE_ON_ERR);
|
2011-08-04 20:01:00 +08:00
|
|
|
} else {
|
2022-10-30 19:55:06 +08:00
|
|
|
struct child_process cmd = CHILD_PROCESS_INIT;
|
2022-10-30 19:47:02 +08:00
|
|
|
|
2022-10-30 19:55:06 +08:00
|
|
|
cmd.git_cmd = 1;
|
|
|
|
strvec_pushl(&cmd.args, "checkout", "-q",
|
|
|
|
oid_to_hex(bisect_rev), "--", NULL);
|
|
|
|
if (run_command(&cmd))
|
2020-02-17 16:40:35 +08:00
|
|
|
/*
|
|
|
|
* Errors in `run_command()` itself, signaled by res < 0,
|
|
|
|
* and errors in the child process, signaled by res > 0
|
2021-07-29 01:07:02 +08:00
|
|
|
* can both be treated as regular BISECT_FAILED (-1).
|
2020-02-17 16:40:35 +08:00
|
|
|
*/
|
2021-07-29 01:07:02 +08:00
|
|
|
return BISECT_FAILED;
|
2011-08-04 20:01:00 +08:00
|
|
|
}
|
2009-04-19 17:56:07 +08:00
|
|
|
|
2021-07-28 02:22:18 +08:00
|
|
|
commit = lookup_commit_reference(the_repository, bisect_rev);
|
2023-03-28 21:58:51 +08:00
|
|
|
repo_format_commit_message(the_repository, commit, "[%H] %s%n",
|
|
|
|
&commit_msg, &pp);
|
2021-07-28 02:22:18 +08:00
|
|
|
fputs(commit_msg.buf, stdout);
|
|
|
|
strbuf_release(&commit_msg);
|
|
|
|
|
2021-07-29 01:07:02 +08:00
|
|
|
return BISECT_OK;
|
2009-04-19 17:56:07 +08:00
|
|
|
}
|
|
|
|
|
2018-11-10 13:48:59 +08:00
|
|
|
static struct commit *get_commit_reference(struct repository *r,
|
|
|
|
const struct object_id *oid)
|
2009-05-09 23:55:45 +08:00
|
|
|
{
|
2018-11-10 13:48:59 +08:00
|
|
|
struct commit *c = lookup_commit_reference(r, oid);
|
|
|
|
if (!c)
|
2017-03-27 00:01:37 +08:00
|
|
|
die(_("Not a valid commit name %s"), oid_to_hex(oid));
|
2018-11-10 13:48:59 +08:00
|
|
|
return c;
|
2009-05-09 23:55:45 +08:00
|
|
|
}
|
|
|
|
|
2018-11-10 13:48:59 +08:00
|
|
|
static struct commit **get_bad_and_good_commits(struct repository *r,
|
|
|
|
int *rev_nr)
|
2009-05-09 23:55:45 +08:00
|
|
|
{
|
2016-02-23 06:44:25 +08:00
|
|
|
struct commit **rev;
|
2009-05-09 23:55:45 +08:00
|
|
|
int i, n = 0;
|
|
|
|
|
2016-02-23 06:44:25 +08:00
|
|
|
ALLOC_ARRAY(rev, 1 + good_revs.nr);
|
2018-11-10 13:48:59 +08:00
|
|
|
rev[n++] = get_commit_reference(r, current_bad_oid);
|
2011-05-20 05:34:33 +08:00
|
|
|
for (i = 0; i < good_revs.nr; i++)
|
2018-11-10 13:48:59 +08:00
|
|
|
rev[n++] = get_commit_reference(r, good_revs.oid + i);
|
2009-05-09 23:55:45 +08:00
|
|
|
*rev_nr = n;
|
|
|
|
|
|
|
|
return rev;
|
|
|
|
}
|
|
|
|
|
2020-02-17 16:40:38 +08:00
|
|
|
static enum bisect_error handle_bad_merge_base(void)
|
2009-05-09 23:55:45 +08:00
|
|
|
{
|
2015-03-14 07:39:29 +08:00
|
|
|
if (is_expected_rev(current_bad_oid)) {
|
|
|
|
char *bad_hex = oid_to_hex(current_bad_oid);
|
2020-03-30 22:04:06 +08:00
|
|
|
char *good_hex = join_oid_array_hex(&good_revs, ' ');
|
2015-06-29 23:40:29 +08:00
|
|
|
if (!strcmp(term_bad, "bad") && !strcmp(term_good, "good")) {
|
2016-06-18 04:21:12 +08:00
|
|
|
fprintf(stderr, _("The merge base %s is bad.\n"
|
2015-06-29 23:40:29 +08:00
|
|
|
"This means the bug has been fixed "
|
2016-06-18 04:21:12 +08:00
|
|
|
"between %s and [%s].\n"),
|
2015-06-29 23:40:29 +08:00
|
|
|
bad_hex, bad_hex, good_hex);
|
2015-06-29 23:40:33 +08:00
|
|
|
} else if (!strcmp(term_bad, "new") && !strcmp(term_good, "old")) {
|
2016-06-18 04:21:12 +08:00
|
|
|
fprintf(stderr, _("The merge base %s is new.\n"
|
2015-06-29 23:40:33 +08:00
|
|
|
"The property has changed "
|
2016-06-18 04:21:12 +08:00
|
|
|
"between %s and [%s].\n"),
|
2015-06-29 23:40:33 +08:00
|
|
|
bad_hex, bad_hex, good_hex);
|
2015-06-29 23:40:29 +08:00
|
|
|
} else {
|
2016-06-18 04:21:12 +08:00
|
|
|
fprintf(stderr, _("The merge base %s is %s.\n"
|
2015-06-29 23:40:29 +08:00
|
|
|
"This means the first '%s' commit is "
|
2016-06-18 04:21:12 +08:00
|
|
|
"between %s and [%s].\n"),
|
2015-06-29 23:40:29 +08:00
|
|
|
bad_hex, term_bad, term_good, bad_hex, good_hex);
|
|
|
|
}
|
2020-02-17 16:40:38 +08:00
|
|
|
return BISECT_MERGE_BASE_CHECK;
|
2009-05-09 23:55:45 +08:00
|
|
|
}
|
|
|
|
|
2016-12-05 06:04:23 +08:00
|
|
|
fprintf(stderr, _("Some %s revs are not ancestors of the %s rev.\n"
|
2009-05-09 23:55:45 +08:00
|
|
|
"git bisect cannot work properly in this case.\n"
|
2016-06-18 04:21:12 +08:00
|
|
|
"Maybe you mistook %s and %s revs?\n"),
|
2015-06-29 23:40:29 +08:00
|
|
|
term_good, term_bad, term_good, term_bad);
|
2020-02-17 16:40:38 +08:00
|
|
|
return BISECT_FAILED;
|
2009-05-09 23:55:45 +08:00
|
|
|
}
|
|
|
|
|
2017-03-31 09:39:57 +08:00
|
|
|
static void handle_skipped_merge_base(const struct object_id *mb)
|
2009-05-09 23:55:45 +08:00
|
|
|
{
|
2017-03-31 09:39:57 +08:00
|
|
|
char *mb_hex = oid_to_hex(mb);
|
2016-06-25 07:09:22 +08:00
|
|
|
char *bad_hex = oid_to_hex(current_bad_oid);
|
2020-03-30 22:04:06 +08:00
|
|
|
char *good_hex = join_oid_array_hex(&good_revs, ' ');
|
2009-05-09 23:55:45 +08:00
|
|
|
|
2016-06-18 04:21:12 +08:00
|
|
|
warning(_("the merge base between %s and [%s] "
|
2009-05-09 23:55:45 +08:00
|
|
|
"must be skipped.\n"
|
2015-06-29 23:40:29 +08:00
|
|
|
"So we cannot be sure the first %s commit is "
|
2009-05-09 23:55:45 +08:00
|
|
|
"between %s and %s.\n"
|
2016-06-18 04:21:12 +08:00
|
|
|
"We continue anyway."),
|
2015-06-29 23:40:29 +08:00
|
|
|
bad_hex, good_hex, term_bad, mb_hex, bad_hex);
|
2009-05-09 23:55:45 +08:00
|
|
|
free(good_hex);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2015-06-29 23:40:33 +08:00
|
|
|
* "check_merge_bases" checks that merge bases are not "bad" (or "new").
|
2009-05-09 23:55:45 +08:00
|
|
|
*
|
2015-06-29 23:40:33 +08:00
|
|
|
* - If one is "bad" (or "new"), it means the user assumed something wrong
|
2020-02-17 16:40:36 +08:00
|
|
|
* and we must return error with a non 0 error code.
|
2015-06-29 23:40:33 +08:00
|
|
|
* - If one is "good" (or "old"), that's good, we have nothing to do.
|
2009-05-09 23:55:45 +08:00
|
|
|
* - If one is "skipped", we can't know but we should warn.
|
|
|
|
* - If we don't know, we should check it out and ask the user to test.
|
2020-02-17 16:40:36 +08:00
|
|
|
* - If a merge base must be tested, on success return
|
|
|
|
* BISECT_INTERNAL_SUCCESS_MERGE_BASE (-11) a special condition
|
|
|
|
* for early success, this will be converted back to 0 in
|
|
|
|
* check_good_are_ancestors_of_bad().
|
2009-05-09 23:55:45 +08:00
|
|
|
*/
|
2020-02-17 16:40:36 +08:00
|
|
|
static enum bisect_error check_merge_bases(int rev_nr, struct commit **rev, int no_checkout)
|
2009-05-09 23:55:45 +08:00
|
|
|
{
|
2020-02-17 16:40:36 +08:00
|
|
|
enum bisect_error res = BISECT_OK;
|
2024-02-28 17:44:16 +08:00
|
|
|
struct commit_list *result = NULL;
|
2009-05-09 23:55:45 +08:00
|
|
|
|
2024-02-28 17:44:16 +08:00
|
|
|
if (repo_get_merge_bases_many(the_repository, rev[0], rev_nr - 1,
|
|
|
|
rev + 1, &result) < 0)
|
|
|
|
exit(128);
|
2009-05-09 23:55:45 +08:00
|
|
|
|
|
|
|
for (; result; result = result->next) {
|
2017-03-31 09:39:57 +08:00
|
|
|
const struct object_id *mb = &result->item->object.oid;
|
convert "oidcmp() == 0" to oideq()
Using the more restrictive oideq() should, in the long run,
give the compiler more opportunities to optimize these
callsites. For now, this conversion should be a complete
noop with respect to the generated code.
The result is also perhaps a little more readable, as it
avoids the "zero is equal" idiom. Since it's so prevalent in
C, I think seasoned programmers tend not to even notice it
anymore, but it can sometimes make for awkward double
negations (e.g., we can drop a few !!oidcmp() instances
here).
This patch was generated almost entirely by the included
coccinelle patch. This mechanical conversion should be
completely safe, because we check explicitly for cases where
oidcmp() is compared to 0, which is what oideq() is doing
under the hood. Note that we don't have to catch "!oidcmp()"
separately; coccinelle's standard isomorphisms make sure the
two are treated equivalently.
I say "almost" because I did hand-edit the coccinelle output
to fix up a few style violations (it mostly keeps the
original formatting, but sometimes unwraps long lines).
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-08-29 05:22:40 +08:00
|
|
|
if (oideq(mb, current_bad_oid)) {
|
2020-02-17 16:40:38 +08:00
|
|
|
res = handle_bad_merge_base();
|
|
|
|
break;
|
2017-03-31 09:40:00 +08:00
|
|
|
} else if (0 <= oid_array_lookup(&good_revs, mb)) {
|
2009-05-09 23:55:45 +08:00
|
|
|
continue;
|
2017-03-31 09:40:00 +08:00
|
|
|
} else if (0 <= oid_array_lookup(&skipped_revs, mb)) {
|
2009-05-09 23:55:45 +08:00
|
|
|
handle_skipped_merge_base(mb);
|
|
|
|
} else {
|
2016-06-18 04:21:12 +08:00
|
|
|
printf(_("Bisecting: a merge base must be tested\n"));
|
2020-02-17 16:40:36 +08:00
|
|
|
res = bisect_checkout(mb, no_checkout);
|
|
|
|
if (!res)
|
|
|
|
/* indicate early success */
|
|
|
|
res = BISECT_INTERNAL_SUCCESS_MERGE_BASE;
|
|
|
|
break;
|
2009-05-09 23:55:45 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
free_commit_list(result);
|
2020-02-17 16:40:36 +08:00
|
|
|
return res;
|
2009-05-09 23:55:45 +08:00
|
|
|
}
|
|
|
|
|
2018-11-10 13:48:59 +08:00
|
|
|
static int check_ancestors(struct repository *r, int rev_nr,
|
|
|
|
struct commit **rev, const char *prefix)
|
2009-05-09 23:55:46 +08:00
|
|
|
{
|
2022-08-02 23:33:15 +08:00
|
|
|
struct strvec rev_argv = STRVEC_INIT;
|
2009-05-17 23:36:46 +08:00
|
|
|
struct rev_info revs;
|
2011-10-02 00:16:08 +08:00
|
|
|
int res;
|
2009-05-09 23:55:46 +08:00
|
|
|
|
2022-08-02 23:33:15 +08:00
|
|
|
bisect_rev_setup(r, &revs, &rev_argv, prefix, "^%s", "%s", 0);
|
2009-05-09 23:55:46 +08:00
|
|
|
|
2009-05-17 23:36:46 +08:00
|
|
|
bisect_common(&revs);
|
|
|
|
res = (revs.commits != NULL);
|
|
|
|
|
|
|
|
/* Clean up objects used, as they will be reused. */
|
2017-12-26 01:45:36 +08:00
|
|
|
clear_commit_marks_many(rev_nr, rev, ALL_REV_FLAGS);
|
2009-05-09 23:55:46 +08:00
|
|
|
|
2022-04-14 04:01:36 +08:00
|
|
|
release_revisions(&revs);
|
2022-08-02 23:33:15 +08:00
|
|
|
strvec_clear(&rev_argv);
|
2009-05-17 23:36:46 +08:00
|
|
|
return res;
|
2009-05-09 23:55:46 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* "check_good_are_ancestors_of_bad" checks that all "good" revs are
|
|
|
|
* ancestor of the "bad" rev.
|
|
|
|
*
|
|
|
|
* If that's not the case, we need to check the merge bases.
|
|
|
|
* If a merge base must be tested by the user, its source code will be
|
2020-02-17 16:40:37 +08:00
|
|
|
* checked out to be tested by the user and we will return.
|
2009-05-09 23:55:46 +08:00
|
|
|
*/
|
2020-02-17 16:40:37 +08:00
|
|
|
|
|
|
|
static enum bisect_error check_good_are_ancestors_of_bad(struct repository *r,
|
2018-11-10 13:48:59 +08:00
|
|
|
const char *prefix,
|
|
|
|
int no_checkout)
|
2009-05-09 23:55:46 +08:00
|
|
|
{
|
2020-02-17 16:40:37 +08:00
|
|
|
char *filename;
|
2009-05-09 23:55:46 +08:00
|
|
|
struct stat st;
|
2017-12-26 01:45:36 +08:00
|
|
|
int fd, rev_nr;
|
2020-02-17 16:40:36 +08:00
|
|
|
enum bisect_error res = BISECT_OK;
|
2017-12-26 01:45:36 +08:00
|
|
|
struct commit **rev;
|
2009-05-09 23:55:46 +08:00
|
|
|
|
2015-03-14 07:39:29 +08:00
|
|
|
if (!current_bad_oid)
|
2020-02-17 16:40:37 +08:00
|
|
|
return error(_("a %s revision is needed"), term_bad);
|
|
|
|
|
|
|
|
filename = git_pathdup("BISECT_ANCESTORS_OK");
|
2009-05-09 23:55:46 +08:00
|
|
|
|
|
|
|
/* Check if file BISECT_ANCESTORS_OK exists. */
|
|
|
|
if (!stat(filename, &st) && S_ISREG(st.st_mode))
|
2012-04-27 06:26:59 +08:00
|
|
|
goto done;
|
2009-05-09 23:55:46 +08:00
|
|
|
|
|
|
|
/* Bisecting with no good rev is ok. */
|
2020-02-17 16:40:30 +08:00
|
|
|
if (!good_revs.nr)
|
2012-04-27 06:26:59 +08:00
|
|
|
goto done;
|
2009-05-09 23:55:46 +08:00
|
|
|
|
2009-05-17 23:36:46 +08:00
|
|
|
/* Check if all good revs are ancestor of the bad rev. */
|
2020-02-17 16:40:36 +08:00
|
|
|
|
2018-11-10 13:48:59 +08:00
|
|
|
rev = get_bad_and_good_commits(r, &rev_nr);
|
|
|
|
if (check_ancestors(r, rev_nr, rev, prefix))
|
2020-02-17 16:40:36 +08:00
|
|
|
res = check_merge_bases(rev_nr, rev, no_checkout);
|
2017-12-26 01:45:36 +08:00
|
|
|
free(rev);
|
2009-05-09 23:55:46 +08:00
|
|
|
|
2020-02-17 16:40:37 +08:00
|
|
|
if (!res) {
|
|
|
|
/* Create file BISECT_ANCESTORS_OK. */
|
|
|
|
fd = open(filename, O_CREAT | O_TRUNC | O_WRONLY, 0600);
|
|
|
|
if (fd < 0)
|
|
|
|
/*
|
|
|
|
* BISECT_ANCESTORS_OK file is not absolutely necessary,
|
|
|
|
* the bisection process will continue at the next
|
|
|
|
* bisection step.
|
|
|
|
* So, just signal with a warning that something
|
|
|
|
* might be wrong.
|
|
|
|
*/
|
|
|
|
warning_errno(_("could not create file '%s'"),
|
|
|
|
filename);
|
|
|
|
else
|
|
|
|
close(fd);
|
|
|
|
}
|
2012-04-27 06:26:59 +08:00
|
|
|
done:
|
|
|
|
free(filename);
|
2020-02-17 16:40:37 +08:00
|
|
|
return res;
|
2009-05-09 23:55:46 +08:00
|
|
|
}
|
|
|
|
|
2009-05-29 05:21:16 +08:00
|
|
|
/*
|
2024-04-14 04:14:48 +08:00
|
|
|
* Display a commit summary to the user.
|
2009-05-29 05:21:16 +08:00
|
|
|
*/
|
2024-04-14 04:14:48 +08:00
|
|
|
static void show_commit(struct commit *commit)
|
2009-05-29 05:21:16 +08:00
|
|
|
{
|
2024-04-14 04:14:48 +08:00
|
|
|
struct child_process show = CHILD_PROCESS_INIT;
|
2009-05-29 05:21:16 +08:00
|
|
|
|
2024-04-14 04:14:48 +08:00
|
|
|
/*
|
|
|
|
* Call git show with --no-pager, as it would otherwise
|
|
|
|
* paginate the "git show" output only, not the output
|
|
|
|
* from bisect_next_all(); this can be fixed by moving
|
|
|
|
* it into a --format parameter, but that would override
|
|
|
|
* the user's default options for "git show", which we
|
|
|
|
* are trying to honour.
|
|
|
|
*/
|
|
|
|
strvec_pushl(&show.args,
|
|
|
|
"--no-pager",
|
|
|
|
"show",
|
|
|
|
"--stat",
|
|
|
|
"--summary",
|
|
|
|
"--no-abbrev-commit",
|
|
|
|
"--diff-merges=first-parent",
|
|
|
|
oid_to_hex(&commit->object.oid), NULL);
|
|
|
|
show.git_cmd = 1;
|
|
|
|
if (run_command(&show))
|
|
|
|
die(_("unable to start 'show' for object '%s'"),
|
|
|
|
oid_to_hex(&commit->object.oid));
|
2009-05-29 05:21:16 +08:00
|
|
|
}
|
|
|
|
|
2015-06-29 23:40:30 +08:00
|
|
|
/*
|
|
|
|
* The terms used for this bisect session are stored in BISECT_TERMS.
|
|
|
|
* We read them and store them to adapt the messages accordingly.
|
|
|
|
* Default is bad/good.
|
|
|
|
*/
|
|
|
|
void read_bisect_terms(const char **read_bad, const char **read_good)
|
|
|
|
{
|
|
|
|
struct strbuf str = STRBUF_INIT;
|
2017-04-21 05:08:25 +08:00
|
|
|
const char *filename = git_path_bisect_terms();
|
2015-06-29 23:40:30 +08:00
|
|
|
FILE *fp = fopen(filename, "r");
|
|
|
|
|
|
|
|
if (!fp) {
|
|
|
|
if (errno == ENOENT) {
|
|
|
|
*read_bad = "bad";
|
|
|
|
*read_good = "good";
|
|
|
|
return;
|
|
|
|
} else {
|
2016-06-18 04:21:12 +08:00
|
|
|
die_errno(_("could not read file '%s'"), filename);
|
2015-06-29 23:40:30 +08:00
|
|
|
}
|
|
|
|
} else {
|
2016-01-14 07:31:17 +08:00
|
|
|
strbuf_getline_lf(&str, fp);
|
2015-06-29 23:40:30 +08:00
|
|
|
*read_bad = strbuf_detach(&str, NULL);
|
2016-01-14 07:31:17 +08:00
|
|
|
strbuf_getline_lf(&str, fp);
|
2015-06-29 23:40:30 +08:00
|
|
|
*read_good = strbuf_detach(&str, NULL);
|
|
|
|
}
|
|
|
|
strbuf_release(&str);
|
|
|
|
fclose(fp);
|
|
|
|
}
|
|
|
|
|
2009-04-19 17:56:07 +08:00
|
|
|
/*
|
2020-02-17 16:40:39 +08:00
|
|
|
* We use the convention that return BISECT_INTERNAL_SUCCESS_1ST_BAD_FOUND (-10) means
|
2009-04-19 17:56:07 +08:00
|
|
|
* the bisection process finished successfully.
|
2020-02-17 16:40:39 +08:00
|
|
|
* In this case the calling function or command should not turn a
|
|
|
|
* BISECT_INTERNAL_SUCCESS_1ST_BAD_FOUND return code into an error or a non zero exit code.
|
2020-09-24 20:33:40 +08:00
|
|
|
*
|
|
|
|
* Checking BISECT_INTERNAL_SUCCESS_1ST_BAD_FOUND
|
|
|
|
* in bisect_helper::bisect_next() and only transforming it to 0 at
|
|
|
|
* the end of bisect_helper::cmd_bisect__helper() helps bypassing
|
|
|
|
* all the code related to finding a commit to test.
|
2009-04-19 17:56:07 +08:00
|
|
|
*/
|
2020-08-08 05:58:36 +08:00
|
|
|
enum bisect_error bisect_next_all(struct repository *r, const char *prefix)
|
2009-04-19 17:56:07 +08:00
|
|
|
{
|
2022-08-02 23:33:15 +08:00
|
|
|
struct strvec rev_argv = STRVEC_INIT;
|
2022-04-14 04:01:38 +08:00
|
|
|
struct rev_info revs = REV_INFO_INIT;
|
2009-04-19 17:56:07 +08:00
|
|
|
struct commit_list *tried;
|
2010-01-19 23:13:33 +08:00
|
|
|
int reaches = 0, all = 0, nr, steps;
|
2020-02-17 16:40:34 +08:00
|
|
|
enum bisect_error res = BISECT_OK;
|
2017-07-14 07:49:24 +08:00
|
|
|
struct object_id *bisect_rev;
|
2017-02-17 01:07:12 +08:00
|
|
|
char *steps_msg;
|
2020-09-24 20:33:40 +08:00
|
|
|
/*
|
|
|
|
* If no_checkout is non-zero, the bisection process does not
|
|
|
|
* checkout the trial commit but instead simply updates BISECT_HEAD.
|
|
|
|
*/
|
2024-05-07 15:11:53 +08:00
|
|
|
int no_checkout = refs_ref_exists(get_main_ref_store(the_repository),
|
|
|
|
"BISECT_HEAD");
|
2020-08-08 05:58:38 +08:00
|
|
|
unsigned bisect_flags = 0;
|
2009-04-19 17:56:07 +08:00
|
|
|
|
2015-06-29 23:40:30 +08:00
|
|
|
read_bisect_terms(&term_bad, &term_good);
|
2009-05-09 23:55:42 +08:00
|
|
|
if (read_bisect_refs())
|
2016-06-18 04:21:12 +08:00
|
|
|
die(_("reading bisect refs failed"));
|
2009-05-09 23:55:42 +08:00
|
|
|
|
2020-08-08 05:58:38 +08:00
|
|
|
if (file_exists(git_path_bisect_first_parent()))
|
|
|
|
bisect_flags |= FIND_BISECTION_FIRST_PARENT_ONLY;
|
|
|
|
|
|
|
|
if (skipped_revs.nr)
|
|
|
|
bisect_flags |= FIND_BISECTION_ALL;
|
|
|
|
|
2020-02-17 16:40:37 +08:00
|
|
|
res = check_good_are_ancestors_of_bad(r, prefix, no_checkout);
|
|
|
|
if (res)
|
2022-04-14 04:01:38 +08:00
|
|
|
goto cleanup;
|
2009-05-09 23:55:47 +08:00
|
|
|
|
2022-08-02 23:33:15 +08:00
|
|
|
bisect_rev_setup(r, &revs, &rev_argv, prefix, "%s", "^%s", 1);
|
2020-08-08 05:58:38 +08:00
|
|
|
|
|
|
|
revs.first_parent_only = !!(bisect_flags & FIND_BISECTION_FIRST_PARENT_ONLY);
|
2009-05-17 23:36:44 +08:00
|
|
|
revs.limited = 1;
|
2009-05-09 23:55:42 +08:00
|
|
|
|
2009-05-17 23:36:44 +08:00
|
|
|
bisect_common(&revs);
|
2009-04-19 17:56:07 +08:00
|
|
|
|
2020-08-08 05:58:38 +08:00
|
|
|
find_bisection(&revs.commits, &reaches, &all, bisect_flags);
|
2009-06-06 12:41:34 +08:00
|
|
|
revs.commits = managed_skipped(revs.commits, &tried);
|
2009-04-19 17:56:07 +08:00
|
|
|
|
|
|
|
if (!revs.commits) {
|
|
|
|
/*
|
2020-02-17 16:40:39 +08:00
|
|
|
* We should return error here only if the "bad"
|
2009-04-19 17:56:07 +08:00
|
|
|
* commit is also a "skip" commit.
|
|
|
|
*/
|
2020-02-17 16:40:34 +08:00
|
|
|
res = error_if_skipped_commits(tried, NULL);
|
|
|
|
if (res < 0)
|
2022-08-02 23:33:11 +08:00
|
|
|
goto cleanup;
|
2016-06-18 04:21:12 +08:00
|
|
|
printf(_("%s was both %s and %s\n"),
|
2015-06-29 23:40:29 +08:00
|
|
|
oid_to_hex(current_bad_oid),
|
|
|
|
term_good,
|
|
|
|
term_bad);
|
2020-02-17 16:40:39 +08:00
|
|
|
|
2022-04-14 04:01:38 +08:00
|
|
|
res = BISECT_FAILED;
|
|
|
|
goto cleanup;
|
2009-04-19 17:56:07 +08:00
|
|
|
}
|
|
|
|
|
2010-03-01 06:19:09 +08:00
|
|
|
if (!all) {
|
2016-06-18 04:21:12 +08:00
|
|
|
fprintf(stderr, _("No testable commit found.\n"
|
2021-02-24 05:11:32 +08:00
|
|
|
"Maybe you started with bad path arguments?\n"));
|
2020-02-17 16:40:39 +08:00
|
|
|
|
2022-04-14 04:01:38 +08:00
|
|
|
res = BISECT_NO_TESTABLE_COMMIT;
|
|
|
|
goto cleanup;
|
2010-03-01 06:19:09 +08:00
|
|
|
}
|
|
|
|
|
2017-07-14 07:49:24 +08:00
|
|
|
bisect_rev = &revs.commits->item->object.oid;
|
2009-04-19 17:56:07 +08:00
|
|
|
|
convert "oidcmp() == 0" to oideq()
Using the more restrictive oideq() should, in the long run,
give the compiler more opportunities to optimize these
callsites. For now, this conversion should be a complete
noop with respect to the generated code.
The result is also perhaps a little more readable, as it
avoids the "zero is equal" idiom. Since it's so prevalent in
C, I think seasoned programmers tend not to even notice it
anymore, but it can sometimes make for awkward double
negations (e.g., we can drop a few !!oidcmp() instances
here).
This patch was generated almost entirely by the included
coccinelle patch. This mechanical conversion should be
completely safe, because we check explicitly for cases where
oidcmp() is compared to 0, which is what oideq() is doing
under the hood. Note that we don't have to catch "!oidcmp()"
separately; coccinelle's standard isomorphisms make sure the
two are treated equivalently.
I say "almost" because I did hand-edit the coccinelle output
to fix up a few style violations (it mostly keeps the
original formatting, but sometimes unwraps long lines).
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-08-29 05:22:40 +08:00
|
|
|
if (oideq(bisect_rev, current_bad_oid)) {
|
2020-02-17 16:40:34 +08:00
|
|
|
res = error_if_skipped_commits(tried, current_bad_oid);
|
|
|
|
if (res)
|
2020-02-17 16:40:39 +08:00
|
|
|
return res;
|
2017-07-14 07:49:24 +08:00
|
|
|
printf("%s is the first %s commit\n", oid_to_hex(bisect_rev),
|
2015-06-29 23:40:29 +08:00
|
|
|
term_bad);
|
2020-02-17 16:40:39 +08:00
|
|
|
|
2024-04-14 04:14:48 +08:00
|
|
|
show_commit(revs.commits->item);
|
2020-02-17 16:40:39 +08:00
|
|
|
/*
|
|
|
|
* This means the bisection process succeeded.
|
|
|
|
* Using BISECT_INTERNAL_SUCCESS_1ST_BAD_FOUND (-10)
|
|
|
|
* so that the call chain can simply check
|
|
|
|
* for negative return values for early returns up
|
|
|
|
* until the cmd_bisect__helper() caller.
|
|
|
|
*/
|
2022-04-14 04:01:38 +08:00
|
|
|
res = BISECT_INTERNAL_SUCCESS_1ST_BAD_FOUND;
|
|
|
|
goto cleanup;
|
2009-04-19 17:56:07 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
nr = all - reaches - 1;
|
2010-01-19 23:13:33 +08:00
|
|
|
steps = estimate_bisect_steps(all);
|
2017-02-17 01:07:12 +08:00
|
|
|
|
|
|
|
steps_msg = xstrfmt(Q_("(roughly %d step)", "(roughly %d steps)",
|
|
|
|
steps), steps);
|
C style: use standard style for "TRANSLATORS" comments
Change all the "TRANSLATORS: [...]" comments in the C code to use the
regular Git coding style, and amend the style guide so that the
example there uses that style.
This custom style was necessary back in 2010 when the gettext support
was initially added, and was subsequently documented in commit
cbcfd4e3ea ("i18n: mention "TRANSLATORS:" marker in
Documentation/CodingGuidelines", 2014-04-18).
GNU xgettext hasn't had the parsing limitation that necessitated this
exception for almost 3 years. Since its 0.19 release on 2014-06-02
it's been able to recognize TRANSLATOR comments in the standard Git
comment syntax[1].
Usually we'd like to keep compatibility with software that's that
young, but in this case literally the only person who needs to be
using a gettext newer than 3 years old is Jiang Xin (the only person
who runs & commits "make pot" results), so I think in this case we can
make an exception.
This xgettext parsing feature was added after a thread on the Git
mailing list[2] which continued on the bug-gettext[3] list, but we
never subsequently changed our style & styleguide, do so.
There are already longstanding changes in git that use the standard
comment style & have their TRANSLATORS comments extracted properly
without getting the literal "*"'s mixed up in the text, as would
happen before xgettext 0.19.
Commit 7ff2683253 ("builtin-am: implement -i/--interactive",
2015-08-04) added one such comment, which in commit df0617bfa7 ("l10n:
git.pot: v2.6.0 round 1 (123 new, 41 removed)", 2015-09-05) got picked
up in the po/git.pot file with the right format, showing that Jiang
already runs a modern xgettext.
The xgettext parser does not handle the sort of non-standard comment
style that I'm amending here in sequencer.c, but that isn't standard
Git comment syntax anyway. With this change to sequencer.c & "make
pot" the comment in the pot file is now correct:
#. TRANSLATORS: %s will be "revert", "cherry-pick" or
-#. * "rebase -i".
+#. "rebase -i".
1. http://git.savannah.gnu.org/cgit/gettext.git/commit/?id=10af7fe6bd
2. <2ce9ec406501d112e032c8208417f8100bed04c6.1397712142.git.worldhello.net@gmail.com>
(https://public-inbox.org/git/2ce9ec406501d112e032c8208417f8100bed04c6.1397712142.git.worldhello.net@gmail.com/)
3. https://lists.gnu.org/archive/html/bug-gettext/2014-04/msg00016.html
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Acked-by: Jiang Xin <worldhello.net@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-05-12 05:20:12 +08:00
|
|
|
/*
|
|
|
|
* TRANSLATORS: the last %s will be replaced with "(roughly %d
|
|
|
|
* steps)" translation.
|
|
|
|
*/
|
2016-06-18 04:21:12 +08:00
|
|
|
printf(Q_("Bisecting: %d revision left to test after this %s\n",
|
|
|
|
"Bisecting: %d revisions left to test after this %s\n",
|
|
|
|
nr), nr, steps_msg);
|
2017-02-17 01:07:12 +08:00
|
|
|
free(steps_msg);
|
2020-09-24 20:33:39 +08:00
|
|
|
/* Clean up objects used, as they will be reused. */
|
2020-10-31 20:47:58 +08:00
|
|
|
repo_clear_commit_marks(r, ALL_REV_FLAGS);
|
2009-04-19 17:56:07 +08:00
|
|
|
|
2022-04-14 04:01:38 +08:00
|
|
|
res = bisect_checkout(bisect_rev, no_checkout);
|
|
|
|
cleanup:
|
|
|
|
release_revisions(&revs);
|
2022-08-02 23:33:15 +08:00
|
|
|
strvec_clear(&rev_argv);
|
2022-04-14 04:01:38 +08:00
|
|
|
return res;
|
2009-04-19 17:56:07 +08:00
|
|
|
}
|
|
|
|
|
2012-10-26 23:53:50 +08:00
|
|
|
static inline int exp2i(int n)
|
|
|
|
{
|
|
|
|
return 1 << n;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Estimate the number of bisect steps left (after the current step)
|
|
|
|
*
|
|
|
|
* For any x between 0 included and 2^n excluded, the probability for
|
|
|
|
* n - 1 steps left looks like:
|
|
|
|
*
|
|
|
|
* P(2^n + x) == (2^n - x) / (2^n + x)
|
|
|
|
*
|
|
|
|
* and P(2^n + x) < 0.5 means 2^n < 3x
|
|
|
|
*/
|
|
|
|
int estimate_bisect_steps(int all)
|
|
|
|
{
|
|
|
|
int n, x, e;
|
|
|
|
|
|
|
|
if (all < 3)
|
|
|
|
return 0;
|
|
|
|
|
wrapper: introduce `log2u()`
We have an implementation of a function that computes the log2 for an
integer. While we could instead use log2(3P), that involves floating
point numbers and is thus needlessly complex and inefficient.
We're about to add a second caller that wants to compute log2 for a
`size_t`. Let's thus move the function into "wrapper.h" such that it
becomes generally available.
While at it, tweak the implementation a bit:
- The parameter is converted from `int` to `uintmax_t`. This
conversion is safe to do in "bisect.c" because we already check that
the argument is positive.
- The return value is an `unsigned`. It cannot ever be negative, so it
is pointless for it to be a signed integer.
- Loop until `!n` instead of requiring that `n > 1` and then subtract
1 from the result and add a special case for `!sz`. This helps
compilers to generate more efficient code.
Compilers recognize the pattern of this function and optimize
accordingly. On GCC 14.2 x86_64:
log2u(unsigned long):
test rdi, rdi
je .L3
bsr rax, rdi
ret
.L3:
mov eax, -1
ret
Clang 18.1 does not yet recognize the pattern, but starts to do so on
Clang trunk x86_64. The code isn't quite as efficient as the one
generated by GCC, but still manages to optimize away the loop:
log2u(unsigned long):
test rdi, rdi
je .LBB0_1
shr rdi
bsr rcx, rdi
mov eax, 127
cmovne rax, rcx
xor eax, -64
add eax, 65
ret
.LBB0_1:
mov eax, -1
ret
The pattern is also recognized on other platforms like ARM64 GCC 14.2.0,
where we end up using `clz`:
log2u(unsigned long):
clz x2, x0
cmp x0, 0
mov w1, 63
sub w0, w1, w2
csinv w0, w0, wzr, ne
ret
Note that we have a similar function `fastlog2()` in the reftable code.
As that codebase is separate from the Git codebase we do not adapt it to
use the new function.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2024-09-04 16:53:00 +08:00
|
|
|
n = log2u(all);
|
2012-10-26 23:53:50 +08:00
|
|
|
e = exp2i(n);
|
|
|
|
x = all - e;
|
|
|
|
|
|
|
|
return (e < 3 * x) ? n : n - 1;
|
|
|
|
}
|
2017-09-29 14:49:39 +08:00
|
|
|
|
2022-08-19 18:08:32 +08:00
|
|
|
static int mark_for_removal(const char *refname,
|
2024-08-09 23:37:50 +08:00
|
|
|
const char *referent UNUSED,
|
2022-08-26 01:09:48 +08:00
|
|
|
const struct object_id *oid UNUSED,
|
|
|
|
int flag UNUSED, void *cb_data)
|
2017-09-29 14:49:39 +08:00
|
|
|
{
|
|
|
|
struct string_list *refs = cb_data;
|
|
|
|
char *ref = xstrfmt("refs/bisect%s", refname);
|
|
|
|
string_list_append(refs, ref);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
int bisect_clean_state(void)
|
|
|
|
{
|
|
|
|
int result = 0;
|
|
|
|
|
|
|
|
/* There may be some refs packed during bisection */
|
|
|
|
struct string_list refs_for_removal = STRING_LIST_INIT_NODUP;
|
2024-05-07 15:11:53 +08:00
|
|
|
refs_for_each_ref_in(get_main_ref_store(the_repository),
|
|
|
|
"refs/bisect", mark_for_removal,
|
|
|
|
(void *) &refs_for_removal);
|
2017-09-29 14:49:39 +08:00
|
|
|
string_list_append(&refs_for_removal, xstrdup("BISECT_HEAD"));
|
2023-12-14 21:37:13 +08:00
|
|
|
string_list_append(&refs_for_removal, xstrdup("BISECT_EXPECTED_REV"));
|
2024-05-07 15:11:53 +08:00
|
|
|
result = refs_delete_refs(get_main_ref_store(the_repository),
|
|
|
|
"bisect: remove", &refs_for_removal,
|
|
|
|
REF_NO_DEREF);
|
2017-09-29 14:49:39 +08:00
|
|
|
refs_for_removal.strdup_strings = 1;
|
|
|
|
string_list_clear(&refs_for_removal, 0);
|
|
|
|
unlink_or_warn(git_path_bisect_ancestors_ok());
|
|
|
|
unlink_or_warn(git_path_bisect_log());
|
|
|
|
unlink_or_warn(git_path_bisect_names());
|
|
|
|
unlink_or_warn(git_path_bisect_run());
|
|
|
|
unlink_or_warn(git_path_bisect_terms());
|
2020-08-08 05:58:37 +08:00
|
|
|
unlink_or_warn(git_path_bisect_first_parent());
|
2017-09-29 14:49:39 +08:00
|
|
|
/*
|
|
|
|
* Cleanup BISECT_START last to support the --no-checkout option
|
|
|
|
* introduced in the commit 4796e823a.
|
|
|
|
*/
|
|
|
|
unlink_or_warn(git_path_bisect_start());
|
|
|
|
|
|
|
|
return result;
|
|
|
|
}
|