git/bisect.c

1185 lines
30 KiB
C
Raw Normal View History

#include "cache.h"
#include "config.h"
#include "commit.h"
#include "diff.h"
#include "revision.h"
#include "refs.h"
#include "list-objects.h"
#include "quote.h"
#include "hash-lookup.h"
#include "run-command.h"
#include "log-tree.h"
#include "bisect.h"
#include "oid-array.h"
#include "strvec.h"
#include "commit-slab.h"
#include "commit-reach.h"
#include "object-store.h"
#include "dir.h"
static struct oid_array good_revs;
static struct oid_array skipped_revs;
static struct object_id *current_bad_oid;
static const char *argv_checkout[] = {"checkout", "-q", NULL, "--", NULL};
static const char *term_bad;
static const char *term_good;
/* Remember to update object flag allocation in object.h */
#define COUNTED (1u<<16)
/*
* This is a truly stupid algorithm, but it's only
* used for bisection, and we just don't care enough.
*
* We care just barely enough to avoid recursing for
* non-merge entries.
*/
static int count_distance(struct commit_list *entry)
{
int nr = 0;
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;
}
}
define_commit_slab(commit_weight, int *);
static struct commit_weight commit_weight;
#define DEBUG_BISECT 0
static inline int weight(struct commit_list *elem)
{
return **commit_weight_at(&commit_weight, elem->item);
}
static inline void weight_set(struct commit_list *elem, int weight)
{
**commit_weight_at(&commit_weight, elem->item) = weight;
}
static int count_interesting_parents(struct commit *commit, unsigned bisect_flags)
{
struct commit_list *p;
int count;
for (count = 0, p = commit->parents; p; p = p->next) {
if (!(p->item->object.flags & UNINTERESTING))
count++;
if (bisect_flags & FIND_BISECTION_FIRST_PARENT_ONLY)
break;
}
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>
2020-11-12 16:19:38 +00:00
static inline int approx_halfway(struct commit_list *p, int nr)
{
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>
2020-11-12 16:19:38 +00:00
int diff;
/*
* 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>
2020-11-12 16:19:38 +00:00
* For small number of commits 2 and 3 are halfway of 5, and
* 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>
2020-11-12 16:19:38 +00:00
diff = 2 * weight(p) - nr;
switch (diff) {
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>
2020-11-12 16:19:38 +00:00
/*
* 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;
return 0;
}
}
static void show_list(const char *debug, int counted, int nr,
struct commit_list *list)
{
struct commit_list *p;
if (!DEBUG_BISECT)
return;
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;
unsigned commit_flags = commit->object.flags;
enum object_type type;
unsigned long size;
char *buf = read_object_file(&commit->object.oid, &type,
&size);
const char *subject_start;
int subject_len;
fprintf(stderr, "%c%c%c ",
(commit_flags & TREESAME) ? ' ' : 'T',
(commit_flags & UNINTERESTING) ? 'U' : ' ',
(commit_flags & COUNTED) ? 'C' : ' ');
if (*commit_weight_at(&commit_weight, p->item))
fprintf(stderr, "%3d", weight(p));
else
fprintf(stderr, "---");
fprintf(stderr, " %.*s", 8, oid_to_hex(&commit->object.oid));
for (pp = commit->parents; pp; pp = pp->next)
fprintf(stderr, " %.*s", 8,
oid_to_hex(&pp->item->object.oid));
subject_len = find_commit_subject(buf, &subject_start);
if (subject_len)
fprintf(stderr, " %.*s", subject_len, subject_start);
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;
unsigned commit_flags = p->item->object.flags;
if (commit_flags & TREESAME)
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 */
return oidcmp(&a->commit->object.oid, &b->commit->object.oid);
}
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));
struct strbuf buf = STRBUF_INIT;
int cnt, i;
for (p = list, cnt = 0; p; p = p->next) {
int distance;
unsigned commit_flags = p->item->object.flags;
if (commit_flags & TREESAME)
continue;
distance = weight(p);
if (nr - distance < distance)
distance = nr - distance;
array[cnt].commit = p->item;
array[cnt].distance = distance;
cnt++;
}
QSORT(array, cnt, compare_commit_dist);
for (p = list, i = 0; i < cnt; i++) {
struct object *obj = &(array[i].commit->object);
strbuf_reset(&buf);
strbuf_addf(&buf, "dist=%d", array[i].distance);
add_name_decoration(DECORATION_NONE, buf.buf, obj);
p->item = array[i].commit;
if (i < cnt - 1)
p = p->next;
}
if (p) {
free_commit_list(p->next);
p->next = NULL;
}
strbuf_release(&buf);
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,
unsigned bisect_flags)
{
int n, counted;
struct commit_list *p;
counted = 0;
for (n = 0, p = list; p; p = p->next) {
struct commit *commit = p->item;
unsigned commit_flags = commit->object.flags;
*commit_weight_at(&commit_weight, p->item) = &weights[n++];
switch (count_interesting_parents(commit, bisect_flags)) {
case 0:
if (!(commit_flags & TREESAME)) {
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;
if (bisect_flags & FIND_BISECTION_FIRST_PARENT_ONLY)
BUG("shouldn't be calling count-distance in fp mode");
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>
2020-11-12 16:19:38 +00:00
/* Does it happen to be at half-way? */
if (!(bisect_flags & FIND_BISECTION_ALL) &&
approx_halfway(p, nr))
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;
unsigned commit_flags = p->item->object.flags;
if (0 <= weight(p))
continue;
for (q = p->item->parents;
q;
q = bisect_flags & FIND_BISECTION_FIRST_PARENT_ONLY ? NULL : q->next) {
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.
*/
if (!(commit_flags & TREESAME)) {
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>
2020-11-12 16:19:38 +00:00
/* Does it happen to be at half-way? */
if (!(bisect_flags & FIND_BISECTION_ALL) &&
approx_halfway(p, nr))
return p;
}
}
show_list("bisection 2 counted all", counted, nr, list);
if (!(bisect_flags & FIND_BISECTION_ALL))
return best_bisection(list, nr);
else
return best_bisection_sorted(list, nr);
}
void find_bisection(struct commit_list **commit_list, int *reaches,
int *all, unsigned bisect_flags)
{
int nr, on_list;
struct commit_list *list, *p, *best, *next, *last;
int *weights;
show_list("bisection 2 entry", 0, 0, *commit_list);
init_commit_weight(&commit_weight);
/*
* Count the number of total and tree-changing items on the
* list, while reversing the list.
*/
for (nr = on_list = 0, last = NULL, p = *commit_list;
p;
p = next) {
unsigned commit_flags = p->item->object.flags;
next = p->next;
if (commit_flags & UNINTERESTING) {
free(p);
continue;
}
p->next = last;
last = p;
if (!(commit_flags & TREESAME))
nr++;
on_list++;
}
list = last;
show_list("bisection 2 sorted", 0, nr, list);
*all = nr;
CALLOC_ARRAY(weights, on_list);
/* Do the real work of finding bisection commit. */
best = do_find_bisection(list, nr, weights, bisect_flags);
if (best) {
if (!(bisect_flags & FIND_BISECTION_ALL)) {
list->item = best->item;
free_commit_list(list->next);
best = list;
best->next = NULL;
}
*reaches = weight(best);
}
free(weights);
*commit_list = best;
clear_commit_weight(&commit_weight);
}
static int register_ref(const char *refname, const struct object_id *oid,
int flags, void *cb_data)
{
struct strbuf good_prefix = STRBUF_INIT;
strbuf_addstr(&good_prefix, term_good);
strbuf_addstr(&good_prefix, "-");
if (!strcmp(refname, term_bad)) {
current_bad_oid = xmalloc(sizeof(*current_bad_oid));
oidcpy(current_bad_oid, oid);
} else if (starts_with(refname, good_prefix.buf)) {
oid_array_append(&good_revs, oid);
} else if (starts_with(refname, "skip-")) {
oid_array_append(&skipped_revs, oid);
}
strbuf_release(&good_prefix);
return 0;
}
static int read_bisect_refs(void)
{
return for_each_ref_in("refs/bisect/", register_ref, NULL);
}
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 09:38:57 +00:00
static GIT_PATH_FUNC(git_path_bisect_names, "BISECT_NAMES")
static GIT_PATH_FUNC(git_path_bisect_expected_rev, "BISECT_EXPECTED_REV")
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")
static GIT_PATH_FUNC(git_path_bisect_terms, "BISECT_TERMS")
static GIT_PATH_FUNC(git_path_bisect_first_parent, "BISECT_FIRST_PARENT")
static GIT_PATH_FUNC(git_path_head_name, "head-name")
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 09:38:57 +00:00
static void read_bisect_paths(struct strvec *array)
{
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 09:38:57 +00:00
const char *filename = git_path_bisect_names();
FILE *fp = xfopen(filename, "r");
while (strbuf_getline_lf(&str, fp) != EOF) {
strbuf_trim(&str);
if (sq_dequote_to_strvec(str.buf, array))
die(_("Badly quoted content in file '%s': %s"),
filename, str.buf);
}
strbuf_release(&str);
fclose(fp);
}
static char *join_oid_array_hex(struct oid_array *array, char delim)
{
struct strbuf joined_hexs = STRBUF_INIT;
int i;
for (i = 0; i < array->nr; i++) {
strbuf_addstr(&joined_hexs, oid_to_hex(array->oid + i));
if (i + 1 < array->nr)
strbuf_addch(&joined_hexs, delim);
}
return strbuf_detach(&joined_hexs, NULL);
}
/*
* 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.
*/
struct commit_list *filter_skipped(struct commit_list *list,
struct commit_list **tried,
int show_all,
int *count,
int *skipped_first)
{
struct commit_list *filtered = NULL, **f = &filtered;
*tried = NULL;
if (skipped_first)
*skipped_first = 0;
if (count)
*count = 0;
if (!skipped_revs.nr)
return list;
while (list) {
struct commit_list *next = list->next;
list->next = NULL;
if (0 <= oid_array_lookup(&skipped_revs, &list->item->object.oid)) {
if (skipped_first && !*skipped_first)
*skipped_first = 1;
/* Move current to tried list */
*tried = list;
tried = &list->next;
} else {
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;
}
/* Move current to filtered list */
*f = list;
f = &list->next;
if (count)
(*count)++;
}
list = next;
}
if (skipped_first && *skipped_first == -1)
*skipped_first = 0;
return filtered;
}
#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.
*/
static unsigned get_prn(unsigned count)
{
count = count * 1103515245 + 12345;
return (count/65536) % PRN_MODULO;
}
/*
* Custom integer square root from
* https://en.wikipedia.org/wiki/Integer_square_root
*/
static int sqrti(int val)
{
float d, x = val;
if (!val)
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)
{
struct commit_list *cur, *previous;
int prn, index, i;
prn = get_prn(count);
index = (count * prn / PRN_MODULO) * sqrti(prn) / sqrti(PRN_MODULO);
cur = list;
previous = NULL;
for (i = 0; cur; cur = cur->next, i++) {
if (i == index) {
if (!oideq(&cur->item->object.oid, current_bad_oid))
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;
if (!skipped_revs.nr)
return list;
list = filter_skipped(list, tried, 0, &count, &skipped_first);
if (!skipped_first)
return list;
return skip_away(list, count);
}
static void bisect_rev_setup(struct repository *r, struct rev_info *revs,
const char *prefix,
const char *bad_format, const char *good_format,
int read_paths)
{
struct strvec rev_argv = STRVEC_INIT;
int i;
repo_init_revisions(r, revs, prefix);
revs->abbrev = 0;
revs->commit_format = CMIT_FMT_UNSPECIFIED;
/* rev_argv.argv[0] will be ignored by setup_revisions */
strvec_push(&rev_argv, "bisect_rev_setup");
strvec_pushf(&rev_argv, bad_format, oid_to_hex(current_bad_oid));
for (i = 0; i < good_revs.nr; i++)
strvec_pushf(&rev_argv, good_format,
oid_to_hex(good_revs.oid + i));
strvec_push(&rev_argv, "--");
if (read_paths)
read_bisect_paths(&rev_argv);
setup_revisions(rev_argv.nr, rev_argv.v, revs, NULL);
/* XXX leak rev_argv, as "revs" may still be pointing to it */
}
static void bisect_common(struct rev_info *revs)
{
if (prepare_revision_walk(revs))
die("revision walk setup failed");
if (revs->tree_objects)
list-objects: consume sparse tree walk When creating a pack-file using 'git pack-objects --revs' we provide a list of interesting and uninteresting commits. For example, a push operation would make the local topic branch be interesting and the known remote refs as uninteresting. We want to discover the set of new objects to send to the server as a thin pack. We walk these commits until we discover a frontier of commits such that every commit walk starting at interesting commits ends in a root commit or unintersting commit. We then need to discover which non-commit objects are reachable from uninteresting commits. This commit walk is not changing during this series. The mark_edges_uninteresting() method in list-objects.c iterates on the commit list and does the following: * If the commit is UNINTERSTING, then mark its root tree and every object it can reach as UNINTERESTING. * If the commit is interesting, then mark the root tree of every UNINTERSTING parent (and all objects that tree can reach) as UNINTERSTING. At the very end, we repeat the process on every commit directly given to the revision walk from stdin. This helps ensure we properly cover shallow commits that otherwise were not included in the frontier. The logic to recursively follow trees is in the mark_tree_uninteresting() method in revision.c. The algorithm avoids duplicate work by not recursing into trees that are already marked UNINTERSTING. Add a new 'sparse' option to the mark_edges_uninteresting() method that performs this logic in a slightly different way. As we iterate over the commits, we add all of the root trees to an oidset. Then, call mark_trees_uninteresting_sparse() on that oidset. Note that we include interesting trees in this process. The current implementation of mark_trees_unintersting_sparse() will walk the same trees as the old logic, but this will be replaced in a later change. Add a '--sparse' flag in 'git pack-objects' to call this new logic. Add a new test script t/t5322-pack-objects-sparse.sh that tests this option. The tests currently demonstrate that the resulting object list is the same as the old algorithm. This includes a case where both algorithms pack an object that is not needed by a remote due to limits on the explored set of trees. When the sparse algorithm is changed in a later commit, we will add a test that demonstrates a change of behavior in some cases. Signed-off-by: Derrick Stolee <dstolee@microsoft.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-01-16 18:25:58 +00:00
mark_edges_uninteresting(revs, NULL, 0);
}
static enum bisect_error error_if_skipped_commits(struct commit_list *tried,
const struct object_id *bad)
{
if (!tried)
return BISECT_OK;
printf("There are only 'skip'ped commits left to test.\n"
"The first %s commit could be any of:\n", term_bad);
for ( ; tried; tried = tried->next)
printf("%s\n", oid_to_hex(&tried->item->object.oid));
if (bad)
printf("%s\n", oid_to_hex(bad));
printf(_("We cannot bisect more!\n"));
return BISECT_ONLY_SKIPPED_LEFT;
}
static int is_expected_rev(const struct object_id *oid)
{
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 09:38:57 +00:00
const char *filename = git_path_bisect_expected_rev();
struct stat st;
struct strbuf str = STRBUF_INIT;
FILE *fp;
int res = 0;
if (stat(filename, &st) || !S_ISREG(st.st_mode))
return 0;
fp = fopen_or_warn(filename, "r");
if (!fp)
return 0;
if (strbuf_getline_lf(&str, fp) != EOF)
res = !strcmp(str.buf, oid_to_hex(oid));
strbuf_release(&str);
fclose(fp);
return res;
}
static enum bisect_error bisect_checkout(const struct object_id *bisect_rev, int no_checkout)
{
char bisect_rev_hex[GIT_MAX_HEXSZ + 1];
struct commit *commit;
struct pretty_print_context pp = {0};
struct strbuf commit_msg = STRBUF_INIT;
oid_to_hex_r(bisect_rev_hex, bisect_rev);
update_ref(NULL, "BISECT_EXPECTED_REV", bisect_rev, NULL, 0, UPDATE_REFS_DIE_ON_ERR);
argv_checkout[2] = bisect_rev_hex;
if (no_checkout) {
update_ref(NULL, "BISECT_HEAD", bisect_rev, NULL, 0,
UPDATE_REFS_DIE_ON_ERR);
} else {
if (run_command_v_opt(argv_checkout, RUN_GIT_CMD))
/*
* Errors in `run_command()` itself, signaled by res < 0,
* and errors in the child process, signaled by res > 0
* can both be treated as regular BISECT_FAILED (-1).
*/
return BISECT_FAILED;
}
commit = lookup_commit_reference(the_repository, bisect_rev);
format_commit_message(commit, "[%H] %s%n", &commit_msg, &pp);
fputs(commit_msg.buf, stdout);
strbuf_release(&commit_msg);
return BISECT_OK;
}
static struct commit *get_commit_reference(struct repository *r,
const struct object_id *oid)
{
struct commit *c = lookup_commit_reference(r, oid);
if (!c)
die(_("Not a valid commit name %s"), oid_to_hex(oid));
return c;
}
static struct commit **get_bad_and_good_commits(struct repository *r,
int *rev_nr)
{
struct commit **rev;
int i, n = 0;
ALLOC_ARRAY(rev, 1 + good_revs.nr);
rev[n++] = get_commit_reference(r, current_bad_oid);
for (i = 0; i < good_revs.nr; i++)
rev[n++] = get_commit_reference(r, good_revs.oid + i);
*rev_nr = n;
return rev;
}
static enum bisect_error handle_bad_merge_base(void)
{
if (is_expected_rev(current_bad_oid)) {
char *bad_hex = oid_to_hex(current_bad_oid);
char *good_hex = join_oid_array_hex(&good_revs, ' ');
if (!strcmp(term_bad, "bad") && !strcmp(term_good, "good")) {
fprintf(stderr, _("The merge base %s is bad.\n"
"This means the bug has been fixed "
"between %s and [%s].\n"),
bad_hex, bad_hex, good_hex);
bisect: add the terms old/new When not looking for a regression during a bisect but for a fix or a change in another given property, it can be confusing to use 'good' and 'bad'. This patch introduce `git bisect new` and `git bisect old` as an alternative to 'bad' and good': the commits which have a certain property must be marked as `new` and the ones which do not as `old`. The output will be the first commit after the change in the property. During a new/old bisect session you cannot use bad/good commands and vice-versa. Some commands are still not available for old/new: * git rev-list --bisect does not treat the revs/bisect/new and revs/bisect/old-SHA1 files. Old discussions: - http://thread.gmane.org/gmane.comp.version-control.git/86063 introduced bisect fix unfixed to find fix. - http://thread.gmane.org/gmane.comp.version-control.git/182398 discussion around bisect yes/no or old/new. - http://thread.gmane.org/gmane.comp.version-control.git/199758 last discussion and reviews New discussions: - http://thread.gmane.org/gmane.comp.version-control.git/271320 ( v2 1/7-4/7 ) - http://comments.gmane.org/gmane.comp.version-control.git/271343 ( v2 5/7-7/7 ) Signed-off-by: Antoine Delaite <antoine.delaite@ensimag.grenoble-inp.fr> Signed-off-by: Louis Stuber <stuberl@ensimag.grenoble-inp.fr> Signed-off-by: Valentin Duperray <Valentin.Duperray@ensimag.imag.fr> Signed-off-by: Franck Jonas <Franck.Jonas@ensimag.imag.fr> Signed-off-by: Lucien Kong <Lucien.Kong@ensimag.imag.fr> Signed-off-by: Thomas Nguy <Thomas.Nguy@ensimag.imag.fr> Signed-off-by: Huynh Khoi Nguyen Nguyen <Huynh-Khoi-Nguyen.Nguyen@ensimag.imag.fr> Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-06-29 15:40:33 +00:00
} else if (!strcmp(term_bad, "new") && !strcmp(term_good, "old")) {
fprintf(stderr, _("The merge base %s is new.\n"
bisect: add the terms old/new When not looking for a regression during a bisect but for a fix or a change in another given property, it can be confusing to use 'good' and 'bad'. This patch introduce `git bisect new` and `git bisect old` as an alternative to 'bad' and good': the commits which have a certain property must be marked as `new` and the ones which do not as `old`. The output will be the first commit after the change in the property. During a new/old bisect session you cannot use bad/good commands and vice-versa. Some commands are still not available for old/new: * git rev-list --bisect does not treat the revs/bisect/new and revs/bisect/old-SHA1 files. Old discussions: - http://thread.gmane.org/gmane.comp.version-control.git/86063 introduced bisect fix unfixed to find fix. - http://thread.gmane.org/gmane.comp.version-control.git/182398 discussion around bisect yes/no or old/new. - http://thread.gmane.org/gmane.comp.version-control.git/199758 last discussion and reviews New discussions: - http://thread.gmane.org/gmane.comp.version-control.git/271320 ( v2 1/7-4/7 ) - http://comments.gmane.org/gmane.comp.version-control.git/271343 ( v2 5/7-7/7 ) Signed-off-by: Antoine Delaite <antoine.delaite@ensimag.grenoble-inp.fr> Signed-off-by: Louis Stuber <stuberl@ensimag.grenoble-inp.fr> Signed-off-by: Valentin Duperray <Valentin.Duperray@ensimag.imag.fr> Signed-off-by: Franck Jonas <Franck.Jonas@ensimag.imag.fr> Signed-off-by: Lucien Kong <Lucien.Kong@ensimag.imag.fr> Signed-off-by: Thomas Nguy <Thomas.Nguy@ensimag.imag.fr> Signed-off-by: Huynh Khoi Nguyen Nguyen <Huynh-Khoi-Nguyen.Nguyen@ensimag.imag.fr> Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-06-29 15:40:33 +00:00
"The property has changed "
"between %s and [%s].\n"),
bisect: add the terms old/new When not looking for a regression during a bisect but for a fix or a change in another given property, it can be confusing to use 'good' and 'bad'. This patch introduce `git bisect new` and `git bisect old` as an alternative to 'bad' and good': the commits which have a certain property must be marked as `new` and the ones which do not as `old`. The output will be the first commit after the change in the property. During a new/old bisect session you cannot use bad/good commands and vice-versa. Some commands are still not available for old/new: * git rev-list --bisect does not treat the revs/bisect/new and revs/bisect/old-SHA1 files. Old discussions: - http://thread.gmane.org/gmane.comp.version-control.git/86063 introduced bisect fix unfixed to find fix. - http://thread.gmane.org/gmane.comp.version-control.git/182398 discussion around bisect yes/no or old/new. - http://thread.gmane.org/gmane.comp.version-control.git/199758 last discussion and reviews New discussions: - http://thread.gmane.org/gmane.comp.version-control.git/271320 ( v2 1/7-4/7 ) - http://comments.gmane.org/gmane.comp.version-control.git/271343 ( v2 5/7-7/7 ) Signed-off-by: Antoine Delaite <antoine.delaite@ensimag.grenoble-inp.fr> Signed-off-by: Louis Stuber <stuberl@ensimag.grenoble-inp.fr> Signed-off-by: Valentin Duperray <Valentin.Duperray@ensimag.imag.fr> Signed-off-by: Franck Jonas <Franck.Jonas@ensimag.imag.fr> Signed-off-by: Lucien Kong <Lucien.Kong@ensimag.imag.fr> Signed-off-by: Thomas Nguy <Thomas.Nguy@ensimag.imag.fr> Signed-off-by: Huynh Khoi Nguyen Nguyen <Huynh-Khoi-Nguyen.Nguyen@ensimag.imag.fr> Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-06-29 15:40:33 +00:00
bad_hex, bad_hex, good_hex);
} else {
fprintf(stderr, _("The merge base %s is %s.\n"
"This means the first '%s' commit is "
"between %s and [%s].\n"),
bad_hex, term_bad, term_good, bad_hex, good_hex);
}
return BISECT_MERGE_BASE_CHECK;
}
fprintf(stderr, _("Some %s revs are not ancestors of the %s rev.\n"
"git bisect cannot work properly in this case.\n"
"Maybe you mistook %s and %s revs?\n"),
term_good, term_bad, term_good, term_bad);
return BISECT_FAILED;
}
static void handle_skipped_merge_base(const struct object_id *mb)
{
char *mb_hex = oid_to_hex(mb);
char *bad_hex = oid_to_hex(current_bad_oid);
char *good_hex = join_oid_array_hex(&good_revs, ' ');
warning(_("the merge base between %s and [%s] "
"must be skipped.\n"
"So we cannot be sure the first %s commit is "
"between %s and %s.\n"
"We continue anyway."),
bad_hex, good_hex, term_bad, mb_hex, bad_hex);
free(good_hex);
}
/*
bisect: add the terms old/new When not looking for a regression during a bisect but for a fix or a change in another given property, it can be confusing to use 'good' and 'bad'. This patch introduce `git bisect new` and `git bisect old` as an alternative to 'bad' and good': the commits which have a certain property must be marked as `new` and the ones which do not as `old`. The output will be the first commit after the change in the property. During a new/old bisect session you cannot use bad/good commands and vice-versa. Some commands are still not available for old/new: * git rev-list --bisect does not treat the revs/bisect/new and revs/bisect/old-SHA1 files. Old discussions: - http://thread.gmane.org/gmane.comp.version-control.git/86063 introduced bisect fix unfixed to find fix. - http://thread.gmane.org/gmane.comp.version-control.git/182398 discussion around bisect yes/no or old/new. - http://thread.gmane.org/gmane.comp.version-control.git/199758 last discussion and reviews New discussions: - http://thread.gmane.org/gmane.comp.version-control.git/271320 ( v2 1/7-4/7 ) - http://comments.gmane.org/gmane.comp.version-control.git/271343 ( v2 5/7-7/7 ) Signed-off-by: Antoine Delaite <antoine.delaite@ensimag.grenoble-inp.fr> Signed-off-by: Louis Stuber <stuberl@ensimag.grenoble-inp.fr> Signed-off-by: Valentin Duperray <Valentin.Duperray@ensimag.imag.fr> Signed-off-by: Franck Jonas <Franck.Jonas@ensimag.imag.fr> Signed-off-by: Lucien Kong <Lucien.Kong@ensimag.imag.fr> Signed-off-by: Thomas Nguy <Thomas.Nguy@ensimag.imag.fr> Signed-off-by: Huynh Khoi Nguyen Nguyen <Huynh-Khoi-Nguyen.Nguyen@ensimag.imag.fr> Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-06-29 15:40:33 +00:00
* "check_merge_bases" checks that merge bases are not "bad" (or "new").
*
bisect: add the terms old/new When not looking for a regression during a bisect but for a fix or a change in another given property, it can be confusing to use 'good' and 'bad'. This patch introduce `git bisect new` and `git bisect old` as an alternative to 'bad' and good': the commits which have a certain property must be marked as `new` and the ones which do not as `old`. The output will be the first commit after the change in the property. During a new/old bisect session you cannot use bad/good commands and vice-versa. Some commands are still not available for old/new: * git rev-list --bisect does not treat the revs/bisect/new and revs/bisect/old-SHA1 files. Old discussions: - http://thread.gmane.org/gmane.comp.version-control.git/86063 introduced bisect fix unfixed to find fix. - http://thread.gmane.org/gmane.comp.version-control.git/182398 discussion around bisect yes/no or old/new. - http://thread.gmane.org/gmane.comp.version-control.git/199758 last discussion and reviews New discussions: - http://thread.gmane.org/gmane.comp.version-control.git/271320 ( v2 1/7-4/7 ) - http://comments.gmane.org/gmane.comp.version-control.git/271343 ( v2 5/7-7/7 ) Signed-off-by: Antoine Delaite <antoine.delaite@ensimag.grenoble-inp.fr> Signed-off-by: Louis Stuber <stuberl@ensimag.grenoble-inp.fr> Signed-off-by: Valentin Duperray <Valentin.Duperray@ensimag.imag.fr> Signed-off-by: Franck Jonas <Franck.Jonas@ensimag.imag.fr> Signed-off-by: Lucien Kong <Lucien.Kong@ensimag.imag.fr> Signed-off-by: Thomas Nguy <Thomas.Nguy@ensimag.imag.fr> Signed-off-by: Huynh Khoi Nguyen Nguyen <Huynh-Khoi-Nguyen.Nguyen@ensimag.imag.fr> Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-06-29 15:40:33 +00:00
* - If one is "bad" (or "new"), it means the user assumed something wrong
* and we must return error with a non 0 error code.
bisect: add the terms old/new When not looking for a regression during a bisect but for a fix or a change in another given property, it can be confusing to use 'good' and 'bad'. This patch introduce `git bisect new` and `git bisect old` as an alternative to 'bad' and good': the commits which have a certain property must be marked as `new` and the ones which do not as `old`. The output will be the first commit after the change in the property. During a new/old bisect session you cannot use bad/good commands and vice-versa. Some commands are still not available for old/new: * git rev-list --bisect does not treat the revs/bisect/new and revs/bisect/old-SHA1 files. Old discussions: - http://thread.gmane.org/gmane.comp.version-control.git/86063 introduced bisect fix unfixed to find fix. - http://thread.gmane.org/gmane.comp.version-control.git/182398 discussion around bisect yes/no or old/new. - http://thread.gmane.org/gmane.comp.version-control.git/199758 last discussion and reviews New discussions: - http://thread.gmane.org/gmane.comp.version-control.git/271320 ( v2 1/7-4/7 ) - http://comments.gmane.org/gmane.comp.version-control.git/271343 ( v2 5/7-7/7 ) Signed-off-by: Antoine Delaite <antoine.delaite@ensimag.grenoble-inp.fr> Signed-off-by: Louis Stuber <stuberl@ensimag.grenoble-inp.fr> Signed-off-by: Valentin Duperray <Valentin.Duperray@ensimag.imag.fr> Signed-off-by: Franck Jonas <Franck.Jonas@ensimag.imag.fr> Signed-off-by: Lucien Kong <Lucien.Kong@ensimag.imag.fr> Signed-off-by: Thomas Nguy <Thomas.Nguy@ensimag.imag.fr> Signed-off-by: Huynh Khoi Nguyen Nguyen <Huynh-Khoi-Nguyen.Nguyen@ensimag.imag.fr> Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-06-29 15:40:33 +00:00
* - If one is "good" (or "old"), that's good, we have nothing to do.
* - 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.
* - 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().
*/
static enum bisect_error check_merge_bases(int rev_nr, struct commit **rev, int no_checkout)
{
enum bisect_error res = BISECT_OK;
struct commit_list *result;
result = get_merge_bases_many(rev[0], rev_nr - 1, rev + 1);
for (; result; result = result->next) {
const struct object_id *mb = &result->item->object.oid;
if (oideq(mb, current_bad_oid)) {
res = handle_bad_merge_base();
break;
} else if (0 <= oid_array_lookup(&good_revs, mb)) {
continue;
} else if (0 <= oid_array_lookup(&skipped_revs, mb)) {
handle_skipped_merge_base(mb);
} else {
printf(_("Bisecting: a merge base must be tested\n"));
res = bisect_checkout(mb, no_checkout);
if (!res)
/* indicate early success */
res = BISECT_INTERNAL_SUCCESS_MERGE_BASE;
break;
}
}
free_commit_list(result);
return res;
}
static int check_ancestors(struct repository *r, int rev_nr,
struct commit **rev, const char *prefix)
{
struct rev_info revs;
int res;
bisect_rev_setup(r, &revs, prefix, "^%s", "%s", 0);
bisect_common(&revs);
res = (revs.commits != NULL);
/* Clean up objects used, as they will be reused. */
clear_commit_marks_many(rev_nr, rev, ALL_REV_FLAGS);
return res;
}
/*
* "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
* checked out to be tested by the user and we will return.
*/
static enum bisect_error check_good_are_ancestors_of_bad(struct repository *r,
const char *prefix,
int no_checkout)
{
char *filename;
struct stat st;
int fd, rev_nr;
enum bisect_error res = BISECT_OK;
struct commit **rev;
if (!current_bad_oid)
return error(_("a %s revision is needed"), term_bad);
filename = git_pathdup("BISECT_ANCESTORS_OK");
/* Check if file BISECT_ANCESTORS_OK exists. */
if (!stat(filename, &st) && S_ISREG(st.st_mode))
goto done;
/* Bisecting with no good rev is ok. */
if (!good_revs.nr)
goto done;
/* Check if all good revs are ancestor of the bad rev. */
rev = get_bad_and_good_commits(r, &rev_nr);
if (check_ancestors(r, rev_nr, rev, prefix))
res = check_merge_bases(rev_nr, rev, no_checkout);
free(rev);
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);
}
done:
free(filename);
return res;
}
/*
* This does "git diff-tree --pretty COMMIT" without one fork+exec.
*/
static void show_diff_tree(struct repository *r,
const char *prefix,
struct commit *commit)
{
const char *argv[] = {
bisect: make diff-tree output prettier After completing a bisection, we print out the commit we found using an internal version of diff-tree. The result is aesthetically lacking: - it shows a raw diff, which is generally less informative for human readers than "--stat --summary" (which we already decided was nice for humans in format-patch's output). - by not abbreviating hashes, the result is likely to wrap on most people's terminals - we don't use "-r", so if the commit touched files in a directory, you only get to see the top-level directory mentioned - we don't specify "--cc" or similar, so merges print nothing (not even the commit message!) Even though bisect might be driven by scripts, there's no reason to consider this part of the output as machine-readable (if anything, the initial "$hash is the first bad commit" might be parsed, but we won't touch that here). Let's make it prettier and more informative for a human reading the output. While we're tweaking the options, let's also switch to using the diff "ui" config. If we're accepting that this is human-readable output, then we should respect the user's options for how to display it. Note that we have to touch a few tests in t6030. These check bisection in a corrupted repository (it's missing a subtree). They didn't fail with the previous code, because it didn't actually recurse far enough in the diff to find the broken tree. But now we'll see the corruption and complain. Adjusting the tests to expect the die() is the best fix. We still confirm that we're able to bisect within the broken repo. And we'll still print "$hash is the first bad commit" as usual before dying; showing that is a reasonable outcome in a corrupt repository (and was what might happen already, if the root tree was corrupt). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-02-22 06:23:28 +00:00
"diff-tree", "--pretty", "--stat", "--summary", "--cc", NULL
};
struct rev_info opt;
bisect: make diff-tree output prettier After completing a bisection, we print out the commit we found using an internal version of diff-tree. The result is aesthetically lacking: - it shows a raw diff, which is generally less informative for human readers than "--stat --summary" (which we already decided was nice for humans in format-patch's output). - by not abbreviating hashes, the result is likely to wrap on most people's terminals - we don't use "-r", so if the commit touched files in a directory, you only get to see the top-level directory mentioned - we don't specify "--cc" or similar, so merges print nothing (not even the commit message!) Even though bisect might be driven by scripts, there's no reason to consider this part of the output as machine-readable (if anything, the initial "$hash is the first bad commit" might be parsed, but we won't touch that here). Let's make it prettier and more informative for a human reading the output. While we're tweaking the options, let's also switch to using the diff "ui" config. If we're accepting that this is human-readable output, then we should respect the user's options for how to display it. Note that we have to touch a few tests in t6030. These check bisection in a corrupted repository (it's missing a subtree). They didn't fail with the previous code, because it didn't actually recurse far enough in the diff to find the broken tree. But now we'll see the corruption and complain. Adjusting the tests to expect the die() is the best fix. We still confirm that we're able to bisect within the broken repo. And we'll still print "$hash is the first bad commit" as usual before dying; showing that is a reasonable outcome in a corrupt repository (and was what might happen already, if the root tree was corrupt). Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-02-22 06:23:28 +00:00
git_config(git_diff_ui_config, NULL);
repo_init_revisions(r, &opt, prefix);
setup_revisions(ARRAY_SIZE(argv) - 1, argv, &opt, NULL);
log_tree_commit(&opt, commit);
}
/*
* 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;
const char *filename = git_path_bisect_terms();
FILE *fp = fopen(filename, "r");
if (!fp) {
if (errno == ENOENT) {
*read_bad = "bad";
*read_good = "good";
return;
} else {
die_errno(_("could not read file '%s'"), filename);
}
} else {
strbuf_getline_lf(&str, fp);
*read_bad = strbuf_detach(&str, NULL);
strbuf_getline_lf(&str, fp);
*read_good = strbuf_detach(&str, NULL);
}
strbuf_release(&str);
fclose(fp);
}
/*
* We use the convention that return BISECT_INTERNAL_SUCCESS_1ST_BAD_FOUND (-10) means
* the bisection process finished successfully.
* 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.
*
* 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.
*/
enum bisect_error bisect_next_all(struct repository *r, const char *prefix)
{
struct rev_info revs;
struct commit_list *tried;
int reaches = 0, all = 0, nr, steps;
enum bisect_error res = BISECT_OK;
struct object_id *bisect_rev;
char *steps_msg;
/*
* If no_checkout is non-zero, the bisection process does not
* checkout the trial commit but instead simply updates BISECT_HEAD.
*/
int no_checkout = ref_exists("BISECT_HEAD");
unsigned bisect_flags = 0;
read_bisect_terms(&term_bad, &term_good);
if (read_bisect_refs())
die(_("reading bisect refs failed"));
if (file_exists(git_path_bisect_first_parent()))
bisect_flags |= FIND_BISECTION_FIRST_PARENT_ONLY;
if (skipped_revs.nr)
bisect_flags |= FIND_BISECTION_ALL;
res = check_good_are_ancestors_of_bad(r, prefix, no_checkout);
if (res)
return res;
bisect_rev_setup(r, &revs, prefix, "%s", "^%s", 1);
revs.first_parent_only = !!(bisect_flags & FIND_BISECTION_FIRST_PARENT_ONLY);
revs.limited = 1;
bisect_common(&revs);
find_bisection(&revs.commits, &reaches, &all, bisect_flags);
revs.commits = managed_skipped(revs.commits, &tried);
if (!revs.commits) {
/*
* We should return error here only if the "bad"
* commit is also a "skip" commit.
*/
res = error_if_skipped_commits(tried, NULL);
if (res < 0)
return res;
printf(_("%s was both %s and %s\n"),
oid_to_hex(current_bad_oid),
term_good,
term_bad);
return BISECT_FAILED;
}
if (!all) {
fprintf(stderr, _("No testable commit found.\n"
"Maybe you started with bad path arguments?\n"));
return BISECT_NO_TESTABLE_COMMIT;
}
bisect_rev = &revs.commits->item->object.oid;
if (oideq(bisect_rev, current_bad_oid)) {
res = error_if_skipped_commits(tried, current_bad_oid);
if (res)
return res;
printf("%s is the first %s commit\n", oid_to_hex(bisect_rev),
term_bad);
show_diff_tree(r, prefix, revs.commits->item);
/*
* 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.
*/
return BISECT_INTERNAL_SUCCESS_1ST_BAD_FOUND;
}
nr = all - reaches - 1;
steps = estimate_bisect_steps(all);
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>
2017-05-11 21:20:12 +00:00
/*
* TRANSLATORS: the last %s will be replaced with "(roughly %d
* steps)" translation.
*/
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);
free(steps_msg);
/* Clean up objects used, as they will be reused. */
repo_clear_commit_marks(r, ALL_REV_FLAGS);
return bisect_checkout(bisect_rev, no_checkout);
}
static inline int log2i(int n)
{
int log2 = 0;
for (; n > 1; n >>= 1)
log2++;
return log2;
}
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;
n = log2i(all);
e = exp2i(n);
x = all - e;
return (e < 3 * x) ? n : n - 1;
}
static int mark_for_removal(const char *refname, const struct object_id *oid,
int flag, void *cb_data)
{
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;
for_each_ref_in("refs/bisect", mark_for_removal, (void *) &refs_for_removal);
string_list_append(&refs_for_removal, xstrdup("BISECT_HEAD"));
result = delete_refs("bisect: remove", &refs_for_removal, REF_NO_DEREF);
refs_for_removal.strdup_strings = 1;
string_list_clear(&refs_for_removal, 0);
unlink_or_warn(git_path_bisect_expected_rev());
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());
unlink_or_warn(git_path_bisect_first_parent());
/* Cleanup head-name if it got left by an old version of git-bisect */
unlink_or_warn(git_path_head_name());
/*
* Cleanup BISECT_START last to support the --no-checkout option
* introduced in the commit 4796e823a.
*/
unlink_or_warn(git_path_bisect_start());
return result;
}