commit: avoid scanning trailing comments when 'core.commentChar' is "auto"
PATCH 1/1 commit: avoid scanning trailing comments when 'core.commentChar' is "auto"
When core.commentChar is set to "auto", Git selects a comment character by scanning the commit message contents and avoiding any character already present in the message.
If the message still contains old conflict comments (starting with a comment character), Git assumes that character is in use and chooses a different one. As a result, those existing comment lines are no longer recognized as comments and end up being included in the final commit message.
To avoid this, skip scanning the trailing comment block when selecting the comment character. This allows Git to safely reuse the original character when appropriate, keeping the commit message clean and free of leftover conflict information.
Background:
The "auto" value for core.commentchar was introduced in the commit 84c9dc2 (commit: allow core.commentChar=auto for character auto selection) but did not exhibt this issue at that time.
The bug was introduced in commit a6c2654 (rebase -m: fix --signoff with conflicts) where Git started writing conflict comments to the file at 'rebase_path_message()'.
---
I came across this bug when working on a patch series that removes the global variables related to the "commentChar" config options.
builtin/commit.c | 6 +++++- t/t3418-rebase-continue.sh | 18 ++++++++++++++++++ 2 files changed, 23 insertions(+), 1 deletion(-)
builtin/commit.c +5 -1
@@ -688,6 +688,10 @@ static void adjust_comment_line_char(const struct strbuf *sb)char candidates[] ="#;@!$%^&|:" ;char *candidate;const char *p; +size_t cutoff; + +/* Ignore comment chars in trailing comments (e.g., Conflicts:) */ + cutoff = sb->len -ignored_log_message_bytes (sb->buf, sb->len);if (!memchr (sb->buf, candidates[0 ], sb->len)) {free (comment_line_str_to_free);
@@ -700,7 +704,7 @@ static void adjust_comment_line_char(const struct strbuf *sb) candidate =strchr (candidates, *p);if (candidate) *candidate =' ' ; -for (p = sb->buf; *p; p++) { +for (p = sb->buf; p +1 < sb->buf + cutoff; p++) {if ((p[0 ] =='\n' || p[0 ] =='\r' ) && p[1 ]) { candidate =strchr (candidates, p[1 ]);if (candidate)
t/t3418-rebase-continue.sh +18 -0
@@ -328,6 +328,24 @@ test_expect_success 'there is no --no-reschedule-failed-exec in an ongoing rebas test_expect_code 129 git rebase --edit-todo --no-reschedule-failed-exec ' +test_expect_success 'no change in comment character due to conflicts markers with core.commentChar=auto' ' + test_commit base file && + git checkout -b branch-a && + test_commit A file && + git checkout -b branch-b base && + test_commit B file && + test_must_fail git rebase branch-a && + printf "B\nA\n" >file && + git add file && + write_script fake-editor <<-\EOF && + exit 0 + EOF + FAKE_EDITOR="$(pwd)/fake-editor" && + GIT_EDITOR="\"\$FAKE_EDITOR\"" git -c core.commentChar=auto rebase --continue && + # Check that "#" is still the comment character. + test_grep "# Changes to be committed:" .git/COMMIT_EDITMSG +' + test_orig_head_helper () { test_when_finished 'git rebase --abort && git checkout topic &&
Signature · 1 line
2.49.0
Quoted 15 lines · Ayush Chandekar
When core.commentChar is set to "auto", Git selects a comment character by scanning the commit message contents and avoiding any character already present in the message.
If the message still contains old conflict comments (starting with a comment character), Git assumes that character is in use and chooses a different one. As a result, those existing comment lines are no longer recognized as comments and end up being included in the final commit message.
To avoid this, skip scanning the trailing comment block when selecting the comment character. This allows Git to safely reuse the original character when appropriate, keeping the commit message clean and free of leftover conflict information.
Quoted 5 lines
Background:
The "auto" value for core.commentchar was introduced in the commit
84c9dc2(commit: allow core.commentChar=auto for character auto selection) but did not exhibt this issue at that time.
Use "git log -1 --format=reference", i.e.
84c9dc2c (commit: allow core.commentChar=auto for character auto selection, 2014-05-17)
Quoted 3 lines
The bug was introduced in commit
a6c2654(rebase -m: fix --signoff with conflicts) where Git started writing conflict comments to the file at 'rebase_path_message()'.
Ditto.
Quoted 16 lines
builtin/commit.c +4
-0@@ -688,6 +688,10 @@ static void adjust_comment_line_char(const struct strbuf *sb)char candidates[] ="#;@!$%^&|:" ;char *candidate;const char *p; +size_t cutoff; + +/* Ignore comment chars in trailing comments (e.g., Conflicts:) */ + cutoff = sb->len -ignored_log_message_bytes (sb->buf, sb->len);if (!memchr (sb->buf, candidates[0 ], sb->len)) {free (comment_line_str_to_free);
builtin/commit.c static void adjust_comment_line_char(const struct strbuf *sb) · line 710
Quoted 8 lines
builtin/commit.c +1
-1@@ -700,7 +704,7 @@ static void adjust_comment_line_char(const struct strbuf *sb) candidate =strchr (candidates, *p);if (candidate) *candidate =' ' ; -for (p = sb->buf; *p; p++) { +for (p = sb->buf; p +1 < sb->buf + cutoff; p++) {if ((p[0 ] =='\n' || p[0 ] =='\r' ) && p[1 ]) { candidate =strchr (candidates, p[1 ]);if (candidate)
Looks quite straight-forward. Nice.
Thanks.
Quoted 20 lines · Junio C Hamano
Quoted 17 lines · Ayush Chandekar
When core.commentChar is set to "auto", Git selects a comment character by scanning the commit message contents and avoiding any character already present in the message.
If the message still contains old conflict comments (starting with a comment character), Git assumes that character is in use and chooses a different one. As a result, those existing comment lines are no longer recognized as comments and end up being included in the final commit message.
To avoid this, skip scanning the trailing comment block when selecting the comment character. This allows Git to safely reuse the original character when appropriate, keeping the commit message clean and free of leftover conflict information.
Background:
Quoted 9 lines
Quoted 3 lines
The "auto" value for core.commentchar was introduced in the commit
84c9dc2(commit: allow core.commentChar=auto for character auto selection) but did not exhibt this issue at that time.Use "git log -1 --format=reference", i.e.
84c9dc2c(commit: allow core.commentChar=auto for character auto selection, 2014-05-17)
Got it, Thanks! I'll update it.
Quoted 21 lines · Ayush Chandekar
When core.commentChar is set to "auto", Git selects a comment character by scanning the commit message contents and avoiding any character already present in the message.
If the message still contains old conflict comments (starting with a comment character), Git assumes that character is in use and chooses a different one. As a result, those existing comment lines are no longer recognized as comments and end up being included in the final commit message.
To avoid this, skip scanning the trailing comment block when selecting the comment character. This allows Git to safely reuse the original character when appropriate, keeping the commit message clean and free of leftover conflict information.
Background:
The "auto" value for core.commentchar was introduced in the commit
84c9dc2(commit: allow core.commentChar=auto for character auto selection) but did not exhibt this issue at that time.
Quoted 8 lines
The bug was introduced in commit
a6c2654(rebase -m: fix --signoff with conflicts) where Git started writing conflict comments to the file at 'rebase_path_message()'.Mentored-by Christian Couder <[email protected]>Mentored-by Ghanshyam Thakkar <[email protected]>Signed-off-by Ayush Chandekar <[email protected]>---
Nice explanation.
Quoted 15 lines
t/t3418-rebase-continue.sh +1
-2@@ -328,6 +328,24 @@ test_expect_success 'there is no --no-reschedule-failed-exec in an ongoing rebas test_expect_code 129 git rebase --edit-todo --no-reschedule-failed-exec ' +test_expect_success 'no change in comment character due to conflictsmarkers with core.commentChar=auto' '
t/t3418-rebase-continue.sh +3
-0+ test_commit base file && + git checkout -b branch-a && + test_commit A file &&
t/t3418-rebase-continue.sh test_expect_success 'there is no --no-reschedule-failed-exec in an ongoing rebas · line 344
Quoted 10 lines
t/t3418-rebase-continue.sh +10
-0+ git checkout -b branch-b base && + test_commit B file && + test_must_fail git rebase branch-a && + printf "B\nA\n" >file && + git add file && + write_script fake-editor <<-\EOF && + exit 0 + EOF + FAKE_EDITOR="$(pwd)/fake-editor" && + GIT_EDITOR="\"\$FAKE_EDITOR\"" git -c core.commentChar=auto rebase --continue &&
How about
GIT_EDITOR="cat >actual"
Then you can test_grep on that. Like in
t/t3418-rebase-continue.sh test_expect_success 'there is no --no-reschedule-failed-exec in an ongoing rebas · line 346
Quoted 2 lines
t/t3418-rebase-continue.sh +2
-0+ # Check that "#" is still the comment character. + test_grep "# Changes to be committed:" .git/COMMIT_EDITMSG
Nit:
test_grep "^# Changes to be committed:$"
Quoted 7 lines
+'
t/t3418-rebase-continue.sh +1
-1+ test_orig_head_helper () { test_when_finished 'git rebase --abort && git checkout topic && --2.49.0
Quoted 17 lines · Kristoffer Haugsbakk
Quoted 6 lines
+test_expect_success 'no change in comment character due to conflicts markers with core.commentChar=auto' '
t/t3418-rebase-continue.sh +3
-0+ test_commit base file && + git checkout -b branch-a && + test_commit A file &&t/t3418-rebase-continue.sh test_expect_success 'there is no --no-reschedule-failed-exec in an ongoing rebas · line 344
Quoted 10 lines
t/t3418-rebase-continue.sh +10
-0+ git checkout -b branch-b base && + test_commit B file && + test_must_fail git rebase branch-a && + printf "B\nA\n" >file && + git add file && + write_script fake-editor <<-\EOF && + exit 0 + EOF + FAKE_EDITOR="$(pwd)/fake-editor" && + GIT_EDITOR="\"\$FAKE_EDITOR\"" git -c core.commentChar=auto rebase --continue &&
Quoted 11 lines
How about
GIT_EDITOR="cat >actual"
Then you can
test_grepon that. Like inhttps://lore.kernel.org/git/[email protected]/
Quoted 2 lines
t/t3418-rebase-continue.sh +2
-0+ # Check that "#" is still the comment character. + test_grep "# Changes to be committed:" .git/COMMIT_EDITMSG
Thanks, that's much cleaner and faster!
Quoted 4 lines
Nit:
test_grep"^# Changes to be committed:$"
Thanks for the catch. I'll fix it.
PATCH v2 1/1 commit: avoid scanning trailing comments when 'core.commentChar' is "auto"
In reply to Ayush ChandekarWhen core.commentChar is set to "auto", Git selects a comment character by scanning the commit message contents and avoiding any character already present in the message.
If the message still contains old conflict comments (starting with a comment character), Git assumes that character is in use and chooses a different one. As a result, those existing comment lines are no longer recognized as comments and end up being included in the final commit message.
To avoid this, skip scanning the trailing comment block when selecting the comment character. This allows Git to safely reuse the original character when appropriate, keeping the commit message clean and free of leftover conflict information.
Background:
The "auto" value for core.commentchar was introduced in the commit 84c9dc2c5a (commit: allow core.commentChar=auto for character auto selection, 2014-05-17) but did not exhibt this issue at that time.
The bug was introduced in commit a6c2654f83 (rebase -m: fix --signoff with conflicts, 2024-04-18) where Git started writing conflict comments to the file at 'rebase_path_message()'.
---
Thanks to Christian for mentoring, and to Kristopher and Junio for their reviews!
builtin/commit.c | 6 +++++- t/t3418-rebase-continue.sh | 14 ++++++++++++++ 2 files changed, 19 insertions(+), 1 deletion(-)
builtin/commit.c +5 -1
@@ -688,6 +688,10 @@ static void adjust_comment_line_char(const struct strbuf *sb)char candidates[] ="#;@!$%^&|:" ;char *candidate;const char *p; +size_t cutoff; + +/* Ignore comment chars in trailing comments (e.g., Conflicts:) */ + cutoff = sb->len -ignored_log_message_bytes (sb->buf, sb->len);if (!memchr (sb->buf, candidates[0 ], sb->len)) {free (comment_line_str_to_free);
@@ -700,7 +704,7 @@ static void adjust_comment_line_char(const struct strbuf *sb) candidate =strchr (candidates, *p);if (candidate) *candidate =' ' ; -for (p = sb->buf; *p; p++) { +for (p = sb->buf; p +1 < sb->buf + cutoff; p++) {if ((p[0 ] =='\n' || p[0 ] =='\r' ) && p[1 ]) { candidate =strchr (candidates, p[1 ]);if (candidate)
t/t3418-rebase-continue.sh +14 -0
@@ -328,6 +328,20 @@ test_expect_success 'there is no --no-reschedule-failed-exec in an ongoing rebas test_expect_code 129 git rebase --edit-todo --no-reschedule-failed-exec ' +test_expect_success 'no change in comment character due to conflicts markers with core.commentChar=auto' ' + test_commit base file && + git checkout -b branch-a && + test_commit A file && + git checkout -b branch-b base && + test_commit B file && + test_must_fail git rebase branch-a && + printf "B\nA\n" >file && + git add file && + GIT_EDITOR="cat >actual" git -c core.commentChar=auto rebase --continue && + # Check that "#" is still the comment character. + test_grep "^# Changes to be committed:$" actual +' + test_orig_head_helper () { test_when_finished 'git rebase --abort && git checkout topic &&
Signature · 1 line
2.49.0
Hi Ayush
Quoted 10 lines · Ayush Chandekar
When core.commentChar is set to "auto", Git selects a comment character by scanning the commit message contents and avoiding any character already present in the message.
If the message still contains old conflict comments (starting with a comment character), Git assumes that character is in use and chooses a different one. As a result, those existing comment lines are no longer recognized as comments and end up being included in the final commit message.
Quoted 4 lines
To avoid this, skip scanning the trailing comment block when selecting the comment character. This allows Git to safely reuse the original character when appropriate, keeping the commit message clean and free of leftover conflict information.
This is a good explanation of the problem. Maybe this is another reason to consider removing support for commentChar=auto [1]
[1] https://lore.kernel.org/git/[email protected]/
Quoted 26 lines
Background:
The "auto" value for core.commentchar was introduced in the commit
84c9dc2c5a(commit: allow core.commentChar=auto for character auto selection, 2014-05-17) but did not exhibt this issue at that time.The bug was introduced in commit
a6c2654f83(rebase -m: fix --signoff with conflicts, 2024-04-18) where Git started writing conflict comments to the file at 'rebase_path_message()'.Mentored-by Christian Couder <[email protected]>Mentored-by Ghanshyam Thakkar <[email protected]>Signed-off-by Ayush Chandekar <[email protected]>---
Thanks to Christian for mentoring, and to Kristopher and Junio for their reviews!
builtin/commit.c | 6 +++++- t/t3418-rebase-continue.sh | 14 ++++++++++++++ 2 files changed, 19 insertions(+), 1 deletion(-)builtin/commit.c +0
-0
builtin/commit.c static void adjust_comment_line_char(const struct strbuf *sb) · line 694
Quoted 7 lines
builtin/commit.c +4
-0@@ -688,6 +688,10 @@ static void adjust_comment_line_char(const struct strbuf *sb)char candidates[] ="#;@!$%^&|:" ;char *candidate;const char *p; +size_t cutoff; + +/* Ignore comment chars in trailing comments (e.g., Conflicts:) */ + cutoff = sb->len -ignored_log_message_bytes (sb->buf, sb->len);
This finds the "Conflicts:" line. I was surprised to see that the string it looks for is hard coded and not translated, however the sequencer (also surprisingly) does not translate that message either so it should work.
Quoted 17 lines
builtin/commit.c +1
-1if (!memchr (sb->buf, candidates[0 ], sb->len)) {free (comment_line_str_to_free);@@ -700,7 +704,7 @@ static void adjust_comment_line_char(const struct strbuf *sb) candidate =strchr (candidates, *p);if (candidate) *candidate =' ' ; -for (p = sb->buf; *p; p++) { +for (p = sb->buf; p +1 < sb->buf + cutoff; p++) {if ((p[0 ] =='\n' || p[0 ] =='\r' ) && p[1 ]) { candidate =strchr (candidates, p[1 ]);if (candidate)t/t3418-rebase-continue.sh +0
-0
t/t3418-rebase-continue.sh test_expect_success 'there is no --no-reschedule-failed-exec in an ongoing rebas · line 332
Quoted 5 lines
t/t3418-rebase-continue.sh +2
-0@@ -328,6 +328,20 @@ test_expect_success 'there is no --no-reschedule-failed-exec in an ongoing rebas test_expect_code 129 git rebase --edit-todo --no-reschedule-failed-exec ' +test_expect_success 'no change in comment character due to conflicts markers with core.commentChar=auto' ' + test_commit base file &&
If you used an existing file (F1 or F2) like most of the rest of the tests in this file we could avoid creating this commit and save ourselves a couple of processes.
t/t3418-rebase-continue.sh test_expect_success 'there is no --no-reschedule-failed-exec in an ongoing rebas · line 342
Quoted 10 lines
t/t3418-rebase-continue.sh +10
-0+ git checkout -b branch-a && + test_commit A file && + git checkout -b branch-b base && + test_commit B file && + test_must_fail git rebase branch-a && + printf "B\nA\n" >file && + git add file && + GIT_EDITOR="cat >actual" git -c core.commentChar=auto rebase --continue && + # Check that "#" is still the comment character. + test_grep "^# Changes to be committed:$" actual
I agree that it is a good idea to anchor the start of the message, but I'm not sure it is helpful to anchor the end of the message as we don't want the test to fail just because an unrelated change adds some whitespace to the end of this line. I'd be tempted to drop the ':' for the same reason.
Thanks for fixing this
Phillip
Quoted 5 lines
t/t3418-rebase-continue.sh +2
-0+' + test_orig_head_helper () { test_when_finished 'git rebase --abort && git checkout topic &&
Quoted 9 lines · Phillip Wood
builtin/commit.c static void adjust_comment_line_char(const struct strbuf *sb) · line 694
Quoted 4 lines
builtin/commit.c +4
-0+size_t cutoff; + +/* Ignore comment chars in trailing comments (e.g., Conflicts:) */ + cutoff = sb->len -ignored_log_message_bytes (sb->buf, sb->len);This finds the "Conflicts:" line. I was surprised to see that the string it looks for is hard coded and not translated, however the sequencer (also surprisingly) does not translate that message either so it should work.
There is a funny chicken-and-egg problem, though. It limits the search for "Conflicts" by using wt_status_locate_end() based on the current value of comment_line_str. When core.commentstring is set to "auto", the code that reads the configuration does not touch the comment_line_str variable, which is initialized to '#'. So
[core]
commentstring = '%'
commentstring = auto
would have '%' in comment_line_str upon entering this codepath, let wt_status_locate_end() use '%' as the comment string to find the end of the log message, and then looks for "Conflicts:" in the result.
Which may or may not be what you want.
Quoted 3 lines
If you used an existing file (F1 or F2) like most of the rest of the tests in this file we could avoid creating this commit and save ourselves a couple of processes.
Excellent suggestion.
Quoted 7 lines
t/t3418-rebase-continue.sh test_expect_success 'there is no --no-reschedule-failed-exec in an ongoing rebas · line 342
Quoted 1 line
t/t3418-rebase-continue.sh +1
-0+ test_grep "^# Changes to be committed:$" actualI agree that it is a good idea to anchor the start of the message, but I'm not sure it is helpful to anchor the end of the message as we don't want the test to fail just because an unrelated change adds some whitespace to the end of this line. I'd be tempted to drop the ':' for the same reason.
Again, excellent.
Quoted 3 lines
Thanks for fixing this
Phillip
Thanks.
Quoted 23 lines · Junio C Hamano
Quoted 9 lines · Phillip Wood
builtin/commit.c static void adjust_comment_line_char(const struct strbuf *sb) · line 694
Quoted 4 lines
builtin/commit.c +4
-0+size_t cutoff; + +/* Ignore comment chars in trailing comments (e.g., Conflicts:) */ + cutoff = sb->len -ignored_log_message_bytes (sb->buf, sb->len);This finds the "Conflicts:" line. I was surprised to see that the string it looks for is hard coded and not translated, however the sequencer (also surprisingly) does not translate that message either so it should work.
There is a funny chicken-and-egg problem, though. It limits the search for "Conflicts" by using
wt_status_locate_end()based on the current value ofcomment_line_str. When core.commentstring is set to "auto", the code that reads the configuration does not touch thecomment_line_strvariable, which is initialized to '#'. So[core]
commentstring = '%'
commentstring = auto
Quoted 6 lines
would have '%' in
comment_line_strupon entering this codepath, letwt_status_locate_end()use '%' as the comment string to find the end of the log message, and then looks for "Conflicts:" in the result.Which may or may not be what you want.
This is also being used to append signoff, just before the adjust_comment_line_char() function.
Another thing that could be done is to return the function (adjust_comment_line_char()) when we find a "Conflicts:" line. Because I don't think there's sense in adjusting the comment character when the "Conflicts:" line already has a comment character. But, I would like to have some views on this.
Thanks.
Quoted 22 lines · Junio C Hamano
Quoted 9 lines · Phillip Wood
builtin/commit.c static void adjust_comment_line_char(const struct strbuf *sb) · line 694
Quoted 4 lines
builtin/commit.c +4
-0+size_t cutoff; + +/* Ignore comment chars in trailing comments (e.g., Conflicts:) */ + cutoff = sb->len -ignored_log_message_bytes (sb->buf, sb->len);This finds the "Conflicts:" line. I was surprised to see that the string it looks for is hard coded and not translated, however the sequencer (also surprisingly) does not translate that message either so it should work.
There is a funny chicken-and-egg problem, though. It limits the search for "Conflicts" by using
wt_status_locate_end()based on the current value ofcomment_line_str. When core.commentstring is set to "auto", the code that reads the configuration does not touch thecomment_line_strvariable, which is initialized to '#'. So[core]
commentstring = '%'
commentstring = auto
Quoted 5 lines
would have '%' in
comment_line_strupon entering this codepath, letwt_status_locate_end()use '%' as the comment string to find the end of the log message, and then looks for "Conflicts:" in the result.Which may or may not be what you want.
Oh, good point - I'd not looked at the config parsing. So we'd create conflict comments that look like
% Conflicts:
% some-file.c
but so long as the commit message did not contain a '#' character [1] "git commit" would select '#' as the automatic comment char and our conflicts lines would not be treated as a comment.
Should we be resetting comment_line_str to '#' when core.commentString is set to "auto"? That wont help if the commit message contains a '#' but at least it would be consistently broken.
We could move adjust_comment_line_char() into libgit.a, use that to select the comment character used in append_conflicts_hint() and set core.commentString to that character when we run "git commit". I think doing that would mean that appending conflict comments would always work with core.commentString=auto but it is a more complex solution as we would need to remember the comment character and then pass it to git commit once the user had fixed the conflicts.
One final note - although the commit message mentions a change to "git rebase" I think this problem already affected cherry-pick, revert and merge before that change. In practice I suspect it is only cherry-pick where one is likely to see this problem because the template messages for the other two commands are unlikely to contain a '#'.
Thanks
Phillip
[1] For some reason adjust_comment_line_char() will not select '#' as the comment char if it occurs anywhere in the message but the other candidates are selected so long as they are not the first character on any line.
Quoted 44 lines · Phillip Wood
Quoted 22 lines · Junio C Hamano
Quoted 9 lines · Phillip Wood
builtin/commit.c static void adjust_comment_line_char(const struct strbuf *sb) · line 694
Quoted 4 lines
builtin/commit.c +4
-0+size_t cutoff; + +/* Ignore comment chars in trailing comments (e.g., Conflicts:) */ + cutoff = sb->len -ignored_log_message_bytes (sb->buf, sb->len);This finds the "Conflicts:" line. I was surprised to see that the string it looks for is hard coded and not translated, however the sequencer (also surprisingly) does not translate that message either so it should work.
There is a funny chicken-and-egg problem, though. It limits the search for "Conflicts" by using
wt_status_locate_end()based on the current value ofcomment_line_str. When core.commentstring is set to "auto", the code that reads the configuration does not touch thecomment_line_strvariable, which is initialized to '#'. So[core]
commentstring = '%'
commentstring = autoQuoted 5 lines
would have '%' in
comment_line_strupon entering this codepath, letwt_status_locate_end()use '%' as the comment string to find the end of the log message, and then looks for "Conflicts:" in the result.Which may or may not be what you want.
Oh, good point - I'd not looked at the config parsing. So we'd create conflict comments that look like
% Conflicts:
%some-file.cbut so long as the commit message did not contain a '#' character [1] "git commit" would select '#' as the automatic comment char and our conflicts lines would not be treated as a comment.
Should we be resetting
comment_line_strto '#' when core.commentString is set to "auto"? That wont help if the commit message contains a '#' but at least it would be consistently broken.
Quoted 8 lines
We could move
adjust_comment_line_char()into libgit.a, use that to select the comment character used inappend_conflicts_hint()and set core.commentString to that character when we run "git commit". I think doing that would mean that appending conflict comments would always work with core.commentString=auto but it is a more complex solution as we would need to remember the comment character and then pass it to git commit once the user had fixed the conflicts.
So, my GSoC project is refactoring in order to reduce the global state in Git. I was trying to remove the global variables related to comment characters. What I tried is to create one single function which returns the comment string, and we could then pass a strbuf in case of core.commentString=auto. You can check my attempts on my fork here [2] (check repo_get_comment_line_str() in config.c), and also mentioned this in my blog [3]. I thought I had it figured out, but turns out I failed one test where core.commentString=auto. It was that moment I realised that I would need to remember the comment character or the strbuf in functions. Just wanted to share this in case anything strikes you when looking at the approach.
Quoted 6 lines
One final note - although the commit message mentions a change to "git rebase" I think this problem already affected cherry-pick, revert and merge before that change. In practice I suspect it is only cherry-pick where one is likely to see this problem because the template messages for the other two commands are unlikely to contain a '#'.
Quoted 8 lines
Thanks
Phillip
[1] For some reason
adjust_comment_line_char()will not select '#' as the comment char if it occurs anywhere in the message but the other candidates are selected so long as they are not the first character on any line.
Thanks:)
[2] : https://github.com/ayu-ch/git/commits/comment-line-str-4
[3] : https://ayu-ch.github.io/2025/06/09/gsoc-week-1.html
Hi Ayush
Quoted 12 lines · Ayush Chandekar
So, my GSoC project is refactoring in order to reduce the global state in Git. I was trying to remove the global variables related to comment characters. What I tried is to create one single function which returns the comment string, and we could then pass a strbuf in case of core.commentString=auto. You can check my attempts on my fork here [2] (check
repo_get_comment_line_str()inconfig.c), and also mentioned this in my blog [3]. I thought I had it figured out, but turns out I failed one test where core.commentString=auto. It was that moment I realised that I would need to remember the comment character or the strbuf in functions. Just wanted to share this in case anything strikes you when looking at the approach.
Thanks for that context. I'm not sure about having a single function that handles both cases. There is only one caller that cares about "auto" and the support for that has so many corner cases that don't work I'm putting a patch together to deprecate it and remove it when Git 3.0 is released.
Looking at your code it seems to break the "last one wins" between core.commentChar and core.commentString. Also deferring the parsing until the comment character is used changes the behavior of things like "git -c core.commentchar=$'\n' commit -p". Instead of erroring out straight away it will let the user carefully select what they want to commit and then die which is not very user friendly.
Keeping the current parsing logic and storing the result in struct repository might be a better approach though we should think about how commands that run without a repository will be able to access the system and user config settings.
Thanks
Phillip
Quoted 23 lines · Phillip Wood
Hi Ayush
Quoted 12 lines · Ayush Chandekar
So, my GSoC project is refactoring in order to reduce the global state in Git. I was trying to remove the global variables related to comment characters. What I tried is to create one single function which returns the comment string, and we could then pass a strbuf in case of core.commentString=auto. You can check my attempts on my fork here [2] (check
repo_get_comment_line_str()inconfig.c), and also mentioned this in my blog [3]. I thought I had it figured out, but turns out I failed one test where core.commentString=auto. It was that moment I realised that I would need to remember the comment character or the strbuf in functions. Just wanted to share this in case anything strikes you when looking at the approach.Thanks for that context. I'm not sure about having a single function that handles both cases. There is only one caller that cares about "auto" and the support for that has so many corner cases that don't work I'm putting a patch together to deprecate it and remove it when Git 3.0 is released.
Quoted 7 lines
Looking at your code it seems to break the "last one wins" between core.commentChar and core.commentString. Also deferring the parsing until the comment character is used changes the behavior of things like "git -c core.commentchar=$'\n' commit -p". Instead of erroring out straight away it will let the user carefully select what they want to commit and then die which is not very user friendly.
Thanks for taking a look! I was experimenting with this to see if I could simplify the logic, but I didn't realize upfront how many corner cases it would run into.
Quoted 5 lines
Keeping the current parsing logic and storing the result in struct repository might be a better approach though we should think about how commands that run without a repository will be able to access the system and user config settings.
Yeah, I will follow that approach as I did with other patch series.
Quoted 4 lines
Thanks
Phillip
Thanks a lot!
Ayush:)
Quoted 1 line · Phillip Wood
[1] For some reason
adjust_comment_line_char()will not select '#' as the comment char if it occurs anywhere in the message but the other candidates are selected so long as they are not the first character on any line.
Sorry, forget that - I'd misread the code
Quoted 3 lines · Phillip Wood
Should we be resetting
comment_line_strto '#' when core.commentString is set to "auto"? That wont help if the commit message contains a '#' but at least it would be consistently broken.
Yeah, while I was re-reading the code to parse the configuration file, that was exactly what came to my mind. I offhand did not think of any downside of doing so, but I cannot claim that I have spent enough brain cycles to make sure it is free of bad unintended consequences.
Thanks.
Quoted 20 lines · Phillip Wood
Quoted 20 lines
if (!memchr(sb->buf, candidates[0], sb->len)) { free(comment_line_str_to_free);builtin/commit.c +1
-1@@ -700,7 +704,7 @@ static void adjust_comment_line_char(const struct strbuf *sb) candidate =strchr (candidates, *p);if (candidate) *candidate =' ' ; -for (p = sb->buf; *p; p++) { +for (p = sb->buf; p +1 < sb->buf + cutoff; p++) {if ((p[0 ] =='\n' || p[0 ] =='\r' ) && p[1 ]) { candidate =strchr (candidates, p[1 ]);if (candidate)t/t3418-rebase-continue.sh +0
-0@@ -328,6 +328,20 @@ test_expect_success 'there is no --no-reschedule-failed-exec in an ongoing rebas test_expect_code 129 git rebase --edit-todo --no-reschedule-failed-exec '
Quoted 7 lines
t/t3418-rebase-continue.sh test_expect_success 'there is no --no-reschedule-failed-exec in an ongoing rebas · line 332
Quoted 2 lines
t/t3418-rebase-continue.sh +2
-0+test_expect_success 'no change in comment character due to conflicts markers with core.commentChar=auto' ' + test_commit base file &&If you used an existing file (F1 or F2) like most of the rest of the tests in this file we could avoid creating this commit and save ourselves a couple of processes.
Yeah, right, I'll update it.
Quoted 11 lines
t/t3418-rebase-continue.sh test_expect_success 'there is no --no-reschedule-failed-exec in an ongoing rebas · line 342
Quoted 10 lines
t/t3418-rebase-continue.sh +10
-0+ git checkout -b branch-a && + test_commit A file && + git checkout -b branch-b base && + test_commit B file && + test_must_fail git rebase branch-a && + printf "B\nA\n" >file && + git add file && + GIT_EDITOR="cat >actual" git -c core.commentChar=auto rebase --continue && + # Check that "#" is still the comment character. + test_grep "^# Changes to be committed:$" actual
Quoted 6 lines
I agree that it is a good idea to anchor the start of the message, but I'm not sure it is helpful to anchor the end of the message as we don't want the test to fail just because an unrelated change adds some whitespace to the end of this line. I'd be tempted to drop the ':' for the same reason.
Makes sense, I'll fix it.
Thanks a lot for reviewing!
Quoted 3 lines · Ayush Chandekar
The "auto" value for core.commentchar was introduced in the commit
84c9dc2c5a(commit: allow core.commentChar=auto for character auto selection, 2014-05-17) but did not exhibt this issue at that time.
Nit: s/exhibt/exhibit/
Thanks!
PATCH v3 1/1 commit: avoid scanning trailing comments when 'core.commentChar' is "auto"
In reply to Ayush ChandekarWhen core.commentChar is set to "auto", Git selects a comment character by scanning the commit message contents and avoiding any character already present in the message.
If the message still contains old conflict comments (starting with a comment character), Git assumes that character is in use and chooses a different one. As a result, those existing comment lines are no longer recognized as comments and end up being included in the final commit message.
To avoid this, skip scanning the trailing comment block when selecting the comment character. This allows Git to safely reuse the original character when appropriate, keeping the commit message clean and free of leftover conflict information.
Background:
The "auto" value for core.commentchar was introduced in the commit 84c9dc2c5a (commit: allow core.commentChar=auto for character auto selection, 2014-05-17) but did not exhibit this issue at that time.
The bug was introduced in commit a6c2654f83 (rebase -m: fix --signoff with conflicts, 2024-04-18) where Git started writing conflict comments to the file at 'rebase_path_message()'.
---
Thanks to Christian, Kristoffer, Junio and Phillip for their reviews!
Range-diff with v2:
1: 4e74e7a9a6 ! 1: 693f890a36 commit: avoid scanning trailing comments when 'core.commentChar' is "auto"
@@ Commit message
The "auto" value for core.commentchar was introduced in the commit
84c9dc2c5a (commit: allow core.commentChar=auto for character auto
- selection, 2014-05-17) but did not exhibt this issue at that time.
+ selection, 2014-05-17) but did not exhibit this issue at that time.
The bug was introduced in commit a6c2654f83 (rebase -m: fix --signoff
with conflicts, 2024-04-18) where Git started writing conflict comments
@@ t/t3418-rebase-continue.sh: test_expect_success 'there is no --no-reschedule-fai
'
+test_expect_success 'no change in comment character due to conflicts markers with core.commentChar=auto' '
-+ test_commit base file &&
+ git checkout -b branch-a &&
-+ test_commit A file &&
-+ git checkout -b branch-b base &&
-+ test_commit B file &&
++ test_commit A F1 &&
++ git checkout -b branch-b HEAD^ &&
++ test_commit B F1 &&
+ test_must_fail git rebase branch-a &&
-+ printf "B\nA\n" >file &&
-+ git add file &&
++ printf "B\nA\n" >F1 &&
++ git add F1 &&
+ GIT_EDITOR="cat >actual" git -c core.commentChar=auto rebase --continue &&
+ # Check that "#" is still the comment character.
-+ test_grep "^# Changes to be committed:$" actual
++ test_grep "^# Changes to be committed" actual
+'
+
test_orig_head_helper () {builtin/commit.c | 6 +++++- t/t3418-rebase-continue.sh | 13 +++++++++++++ 2 files changed, 18 insertions(+), 1 deletion(-)
builtin/commit.c +5 -1
@@ -688,6 +688,10 @@ static void adjust_comment_line_char(const struct strbuf *sb)char candidates[] ="#;@!$%^&|:" ;char *candidate;const char *p; +size_t cutoff; + +/* Ignore comment chars in trailing comments (e.g., Conflicts:) */ + cutoff = sb->len -ignored_log_message_bytes (sb->buf, sb->len);if (!memchr (sb->buf, candidates[0 ], sb->len)) {free (comment_line_str_to_free);
@@ -700,7 +704,7 @@ static void adjust_comment_line_char(const struct strbuf *sb) candidate =strchr (candidates, *p);if (candidate) *candidate =' ' ; -for (p = sb->buf; *p; p++) { +for (p = sb->buf; p +1 < sb->buf + cutoff; p++) {if ((p[0 ] =='\n' || p[0 ] =='\r' ) && p[1 ]) { candidate =strchr (candidates, p[1 ]);if (candidate)
t/t3418-rebase-continue.sh +13 -0
@@ -328,6 +328,19 @@ test_expect_success 'there is no --no-reschedule-failed-exec in an ongoing rebas test_expect_code 129 git rebase --edit-todo --no-reschedule-failed-exec ' +test_expect_success 'no change in comment character due to conflicts markers with core.commentChar=auto' ' + git checkout -b branch-a && + test_commit A F1 && + git checkout -b branch-b HEAD^ && + test_commit B F1 && + test_must_fail git rebase branch-a && + printf "B\nA\n" >F1 && + git add F1 && + GIT_EDITOR="cat >actual" git -c core.commentChar=auto rebase --continue && + # Check that "#" is still the comment character. + test_grep "^# Changes to be committed" actual +' + test_orig_head_helper () { test_when_finished 'git rebase --abort && git checkout topic &&
Signature · 1 line
2.49.0
Hi Ayush
Quoted 28 lines · Ayush Chandekar
Range-diff with v2:
1:4e74e7a9a6! 1:693f890a36commit: avoid scanning trailing comments when 'core.commentChar' is "auto"@@ Commit message The "auto" value for core.commentchar was introduced in the commit 84c9dc2c5a (commit: allow core.commentChar=auto for character auto - selection, 2014-05-17) but did not exhibt this issue at that time. + selection, 2014-05-17) but did not exhibit this issue at that time. The bug was introduced in commit a6c2654f83 (rebase -m: fix --signoff with conflicts, 2024-04-18) where Git started writing conflict comments @@ t/t3418-rebase-continue.sh: test_expect_success 'there is no --no-reschedule-fai ' +test_expect_success 'no change in comment character due to conflicts markers with core.commentChar=auto' ' -+ test_commit base file && + git checkout -b branch-a && -+ test_commit A file && -+ git checkout -b branch-b base && -+ test_commit B file && ++ test_commit A F1 && ++ git checkout -b branch-b HEAD^ && ++ test_commit B F1 && + test_must_fail git rebase branch-a && -+ printf "B\nA\n" >file && -+ git add file && ++ printf "B\nA\n" >F1 &&
Quoted 8 lines
++ git add F1 &&
+ GIT_EDITOR="cat >actual" git -c core.commentChar=auto rebase --continue &&
+ # Check that "#" is still the comment character.
-+ test_grep "^# Changes to be committed:$" actual
++ test_grep "^# Changes to be committed" actual
+'
+
test_orig_head_helper () {The changes here look good but I think we want to update the config parsing as well so that comment_line_str is reset to '#' when core.commentString=auto. We probably want to do that in its own commit.
Thanks
Phillip
Quoted 41 lines · Phillip Wood
Hi Ayush
Quoted 28 lines · Ayush Chandekar
Range-diff with v2:
1:4e74e7a9a6! 1:693f890a36commit: avoid scanning trailing comments when 'core.commentChar' is "auto"@@ Commit message The "auto" value for core.commentchar was introduced in the commit 84c9dc2c5a (commit: allow core.commentChar=auto for character auto - selection, 2014-05-17) but did not exhibt this issue at that time. + selection, 2014-05-17) but did not exhibit this issue at that time. The bug was introduced in commit a6c2654f83 (rebase -m: fix --signoff with conflicts, 2024-04-18) where Git started writing conflict comments @@ t/t3418-rebase-continue.sh: test_expect_success 'there is no --no-reschedule-fai ' +test_expect_success 'no change in comment character due to conflicts markers with core.commentChar=auto' ' -+ test_commit base file && + git checkout -b branch-a && -+ test_commit A file && -+ git checkout -b branch-b base && -+ test_commit B file && ++ test_commit A F1 && ++ git checkout -b branch-b HEAD^ && ++ test_commit B F1 && + test_must_fail git rebase branch-a && -+ printf "B\nA\n" >file && -+ git add file && ++ printf "B\nA\n" >F1 &&Quoted 8 lines
++ git add F1 && + GIT_EDITOR="cat >actual" git -c core.commentChar=auto rebase --continue && + # Check that "#" is still the comment character. -+ test_grep "^# Changes to be committed:$" actual ++ test_grep "^# Changes to be committed" actual +' + test_orig_head_helper () {
Quoted 8 lines
The changes here look good but I think we want to update the config parsing as well so that
comment_line_stris reset to '#' when core.commentString=auto. We probably want to do that in its own commit.Thanks
Phillip
maybe something like this?
builtin/commit.c +0 -0
@@ -912,8 +912,10 @@ static int prepare_to_commit(const char
*index_file, const char *prefix,
if (fwrite(sb.buf, 1, sb.len, s->fp) < sb.len)
die_errno(_("could not write commit template"));- if (auto_comment_line_char)
+ if (auto_comment_line_char){
+ comment_line_str = "#";
adjust_comment_line_char(&sb);
+ }
strbuf_release(&sb);or we can do it inside the adjust_comment_line() function.
Thanks!
Ayush
Hi Ayush
Quoted 22 lines · Ayush Chandekar
Quoted 4 lines · Phillip Wood
The changes here look good but I think we want to update the config parsing as well so that
comment_line_stris reset to '#' when core.commentString=auto. We probably want to do that in its own commit.maybe something like this?
builtin/commit.c +3
-1@@ -912,8 +912,10 @@ static int prepare_to_commit(const char *index_file, const char *prefix, if (fwrite(sb.buf, 1, sb.len, s->fp) < sb.len)die_errno (_ ("could not write commit template" )); -if (auto_comment_line_char) +if (auto_comment_line_char){ + comment_line_str ="#" ;adjust_comment_line_char (&sb); + }strbuf_release (&sb);
Quoted 1 line
or we can do it inside the
adjust_comment_line()function.
We need to do it when we parse the config so that append_conflicts_comment() uses '#' as the comment char. See the (whitespace damaged) diff below
Thanks
Phillip
config.c +4 -2
@@ -1537,9 +1537,11 @@ static int git_default_core_config(const char *var, const char *value, !strcmp(var, "core.commentstring")) {if (!value)return config_error_nonbool (var); -else if (!strcasecmp (value,"auto" )) +else if (!strcasecmp (value,"auto" )) { auto_comment_line_char =1 ; -else if (value[0 ]) { +FREE_AND_NULL (comment_line_str_to_free); + comment_line_str ="#" ; + }else if (value[0 ]) {if (strchr (value,'\n' )) return error(_("%s cannot contain
newline"), var);
comment_line_str = value;Hi Phillip,
Quoted 54 lines · Phillip Wood
Hi Ayush
Quoted 22 lines · Ayush Chandekar
Quoted 4 lines · Phillip Wood
The changes here look good but I think we want to update the config parsing as well so that
comment_line_stris reset to '#' when core.commentString=auto. We probably want to do that in its own commit.maybe something like this?
builtin/commit.c +3
-1@@ -912,8 +912,10 @@ static int prepare_to_commit(const char *index_file, const char *prefix, if (fwrite(sb.buf, 1, sb.len, s->fp) < sb.len)die_errno (_ ("could not write commit template" )); -if (auto_comment_line_char) +if (auto_comment_line_char){ + comment_line_str ="#" ;adjust_comment_line_char (&sb); + }strbuf_release (&sb);Quoted 1 line
or we can do it inside the
adjust_comment_line()function.We need to do it when we parse the config so that
append_conflicts_comment()uses '#' as the comment char. See the (whitespace damaged) diff belowThanks
Phillip
config.c +4
-2@@ -1537,9 +1537,11 @@ static int git_default_core_config(const char *var, const char *value, !strcmp(var, "core.commentstring")) {if (!value)return config_error_nonbool (var); -else if (!strcasecmp (value,"auto" )) +else if (!strcasecmp (value,"auto" )) { auto_comment_line_char =1 ; -else if (value[0 ]) { +FREE_AND_NULL (comment_line_str_to_free); + comment_line_str ="#" ; + }else if (value[0 ]) {if (strchr (value,'\n' )) return error(_("%s cannot contain
Quoted 3 lines
newline"), var);
comment_line_str = value;Thanks, I understood it.
What if we simply return the function adjust_comment_line_char() if we get a non-zero value from ignored_log_message_bytes(), i.e we won't scan the commit message in case conflict message exists, and we let the old code exist as it is?
+ if(ignored_log_message_bytes(sb->buf, sb->len))
+ return;
Hi Ayush
Quoted 30 lines · Ayush Chandekar
Quoted 18 lines · Phillip Wood
config.c +4
-2@@ -1537,9 +1537,11 @@ static int git_default_core_config(const char *var, const char *value, !strcmp(var, "core.commentstring")) {if (!value)return config_error_nonbool (var); -else if (!strcasecmp (value,"auto" )) +else if (!strcasecmp (value,"auto" )) { auto_comment_line_char =1 ; -else if (value[0 ]) { +FREE_AND_NULL (comment_line_str_to_free); + comment_line_str ="#" ; + }else if (value[0 ]) {if (strchr (value,'\n' )) return error(_("%s cannot containQuoted 3 lines
newline"), var); comment_line_str = value;Thanks, I understood it.
What if we simply return the function
adjust_comment_line_char()if we get a non-zero value fromignored_log_message_bytes(), i.e we won't scan the commit message in case conflict message exists, and we let the old code exist as it is?
Review
Quoted 2 lines
(quoted diff) +2
-0+ if(ignored_log_message_bytes(sb->buf, sb->len)) + return;
So we'd ignore core.commentChar=auto if we detected conflict comments? That might be surprising to the user - it would mean that we'd always avoid adding the conflict comments to the commit message but we'd lose any lines that begin with the comment string. I think I'm leaning slightly towards the original solution but it is not clear to me that one option is much better that the other.
Thanks
Phillip
Hey, Phillip
Quoted 44 lines · Phillip Wood
Hi Ayush
Quoted 30 lines · Ayush Chandekar
Quoted 18 lines · Phillip Wood
config.c +4
-2@@ -1537,9 +1537,11 @@ static int git_default_core_config(const char *var, const char *value, !strcmp(var, "core.commentstring")) {if (!value)return config_error_nonbool (var); -else if (!strcasecmp (value,"auto" )) +else if (!strcasecmp (value,"auto" )) { auto_comment_line_char =1 ; -else if (value[0 ]) { +FREE_AND_NULL (comment_line_str_to_free); + comment_line_str ="#" ; + }else if (value[0 ]) {if (strchr (value,'\n' )) return error(_("%s cannot containQuoted 3 lines
newline"), var); comment_line_str = value;Thanks, I understood it.
What if we simply return the function
adjust_comment_line_char()if we get a non-zero value fromignored_log_message_bytes(), i.e we won't scan the commit message in case conflict message exists, and we let the old code exist as it is?Review
Quoted 2 lines
(quoted diff) +2
-0+ if(ignored_log_message_bytes(sb->buf, sb->len)) + return;So we'd ignore core.commentChar=auto if we detected conflict comments? That might be surprising to the user - it would mean that we'd always avoid adding the conflict comments to the commit message but we'd lose any lines that begin with the comment string. I think I'm leaning slightly towards the original solution but it is not clear to me that one option is much better that the other.
Quoted 4 lines
Thanks
Phillip
Now that we're planning to get rid of the 'auto' keyword from commentChar [1], do you think it would be better if we just ignored the keyword when we detect conflict comments? Also, how is it that a user will end up having lines starting with the character being the same as the conflict comment's character?
Thanks!
Ayush
Hi Ayush
Quoted 6 lines · Ayush Chandekar
Now that we're planning to get rid of the 'auto' keyword from commentChar [1], do you think it would be better if we just ignored the keyword when we detect conflict comments? Also, how is it that a user will end up having lines starting with the character being the same as the conflict comment's character?
Let see if Junio agrees with depreciation and removing support for commentChar=auto first. If we do deprecate it then we should still support it until it is removed so I'm leaning towards fixing the config parsing to reset comment_line_char to '#' instead.
Thanks
Phillip
cover commit: improve behaviour of core.commentChar=auto for comments in commit messages
In reply to Ayush ChandekarHey everyone,
The aim of this patch series is to improve the behaviour of core.commentChar=auto by the following patches:
1/2 - Fix a bug which reads comment character of the comments in commit message leading to change in the value of comment_line_str and thus resulting the comments in the final commit message. 2/2 - Standardizes the behaviour of code by resetting the 'comment_line_str' to "#" when core.commentChar is set to auto.
Thanks to Junio, Phillip and Kristoffer for reviewing the patches and also Christian for the reviews and mentoring me.
Ayush Chandekar (2):
commit: avoid scanning trailing comments when 'core.commentChar' is
"auto"
config: set comment_line_str to "#" when core.commentChar=auto
builtin/commit.c | 6 +++++- config.c | 6 ++++-- t/t3418-rebase-continue.sh | 13 +++++++++++++ 3 files changed, 22 insertions(+), 3 deletions(-)
Signature · 1 line
2.49.0
PATCH 1/2 commit: avoid scanning trailing comments when 'core.commentChar' is "auto"
When core.commentChar is set to "auto", Git selects a comment character by scanning the commit message contents and avoiding any character already present in the message.
If the message still contains old conflict comments (starting with a comment character), Git assumes that character is in use and chooses a different one. As a result, those existing comment lines are no longer recognized as comments and end up being included in the final commit message.
To avoid this, skip scanning the trailing comment block when selecting the comment character. This allows Git to safely reuse the original character when appropriate, keeping the commit message clean and free of leftover conflict information.
Background:
The "auto" value for core.commentchar was introduced in the commit 84c9dc2c5a (commit: allow core.commentChar=auto for character auto selection, 2014-05-17) but did not exhibit this issue at that time.
The bug was introduced in commit a6c2654f83 (rebase -m: fix --signoff with conflicts, 2024-04-18) where Git started writing conflict comments to the file at 'rebase_path_message()'.
--- builtin/commit.c | 6 +++++- t/t3418-rebase-continue.sh | 13 +++++++++++++ 2 files changed, 18 insertions(+), 1 deletion(-)
builtin/commit.c +5 -1
@@ -688,6 +688,10 @@ static void adjust_comment_line_char(const struct strbuf *sb)char candidates[] ="#;@!$%^&|:" ;char *candidate;const char *p; +size_t cutoff; + +/* Ignore comment chars in trailing comments (e.g., Conflicts:) */ + cutoff = sb->len -ignored_log_message_bytes (sb->buf, sb->len);if (!memchr (sb->buf, candidates[0 ], sb->len)) {free (comment_line_str_to_free);
@@ -700,7 +704,7 @@ static void adjust_comment_line_char(const struct strbuf *sb) candidate =strchr (candidates, *p);if (candidate) *candidate =' ' ; -for (p = sb->buf; *p; p++) { +for (p = sb->buf; p +1 < sb->buf + cutoff; p++) {if ((p[0 ] =='\n' || p[0 ] =='\r' ) && p[1 ]) { candidate =strchr (candidates, p[1 ]);if (candidate)
t/t3418-rebase-continue.sh +13 -0
@@ -328,6 +328,19 @@ test_expect_success 'there is no --no-reschedule-failed-exec in an ongoing rebas test_expect_code 129 git rebase --edit-todo --no-reschedule-failed-exec ' +test_expect_success 'no change in comment character due to conflicts markers with core.commentChar=auto' ' + git checkout -b branch-a && + test_commit A F1 && + git checkout -b branch-b HEAD^ && + test_commit B F1 && + test_must_fail git rebase branch-a && + printf "B\nA\n" >F1 && + git add F1 && + GIT_EDITOR="cat >actual" git -c core.commentChar=auto rebase --continue && + # Check that "#" is still the comment character. + test_grep "^# Changes to be committed" actual +' + test_orig_head_helper () { test_when_finished 'git rebase --abort && git checkout topic &&
Signature · 1 line
2.49.0
PATCH 2/2 config: set comment_line_str to "#" when core.commentChar=auto
In reply to Ayush ChandekarIf conflict comments already use a comment character that isn't "#", and core.commentChar is set "auto", Git will ignore these lines during the scan using ignored_log_message_bytes() and pick a new comment character based on the rest of the message. The newly chosen character may be different from the one used in the conflict comments and therefore, these are no longer treated as comments and end up in the final commit message.
For example, during a rebase if the user previously set core.commentChar=% and then encounters a conflict, conflict comments like "% Conflicts:" are generated. If the user subsequently sets core.commentChar=auto before running rebase --continue, Git parses the "auto" setting and begins scanning. It first uses the existing 'comment_line_str' (which is '%') to detect and ignore conflict comments via ignored_log_message_bytes().
Then, Git scans the rest of the message (excluding conflict comments), sees that none of the remaining lines start with '#' and decides to set comment_line_str to '#'. Since the final commit character differs from the one used in the conflict comments, those lines are no longer considered comments and get included in the final commit message.
Set 'comment_line_str' to '#' when core.commentChar is set to 'auto' to reset any previously set value.
While this does not solve the issue of conflict comment inclusion and the user visible behaviour stays tha same, it standardizes the behaviour of the code by always resetting 'comment_line_str' to '#' when core.commentChar=auto is parsed.
--- config.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
config.c +4 -2
@@ -1537,9 +1537,11 @@ static int git_default_core_config(const char *var, const char *value, !strcmp(var, "core.commentstring")) {if (!value)return config_error_nonbool (var); -else if (!strcasecmp (value,"auto" )) +else if (!strcasecmp (value,"auto" )) { auto_comment_line_char =1 ; -else if (value[0 ]) { +FREE_AND_NULL (comment_line_str_to_free); + comment_line_str ="#" ; + }else if (value[0 ]) {if (strchr (value,'\n' ))return error (_ ("%s cannot contain newline" ), var); comment_line_str = value;
Signature · 1 line
2.49.0
Re: [GSOC PATCH 2/2] config: set comment_line_str to "#" when core.commentChar=auto
Quoted 46 lines · Ayush Chandekar
If conflict comments already use a comment character that isn't "#", and core.commentChar is set "auto", Git will ignore these lines during the scan using
ignored_log_message_bytes()and pick a new comment character based on the rest of the message. The newly chosen character may be different from the one used in the conflict comments and therefore, these are no longer treated as comments and end up in the final commit message.For example, during a rebase if the user previously set core.commentChar=% and then encounters a conflict, conflict comments like "% Conflicts:" are generated. If the user subsequently sets core.commentChar=auto before running
rebase --continue, Git parses the "auto" setting and begins scanning. It first uses the existing 'comment_line_str' (which is '%') to detect and ignore conflict comments viaignored_log_message_bytes().Then, Git scans the rest of the message (excluding conflict comments), sees that none of the remaining lines start with '#' and decides to set
comment_line_strto '#'. Since the final commit character differs from the one used in the conflict comments, those lines are no longer considered comments and get included in the final commit message.Set '
comment_line_str' to '#' when core.commentChar is set to 'auto' to reset any previously set value.While this does not solve the issue of conflict comment inclusion and the user visible behaviour stays tha same, it standardizes the behaviour of the code by always resetting '
comment_line_str' to '#' when core.commentChar=auto is parsed.Mentored-by Christian Couder <[email protected]>Mentored-by Ghanshyam Thakkar <[email protected]>Signed-off-by Ayush Chandekar <[email protected]>--- config.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)config.c +0
-0@@ -1537,9 +1537,11 @@ static int git_default_core_config(const char *var, const char *value, !strcmp(var, "core.commentstring")) {if (!value)return config_error_nonbool (var);
config.c static int git_default_core_config(const char *var, const char *value, · line 1547
Quoted 10 lines
config.c +4
-2-else if (!strcasecmp (value,"auto" )) +else if (!strcasecmp (value,"auto" )) { auto_comment_line_char =1 ; -else if (value[0 ]) { +FREE_AND_NULL (comment_line_str_to_free); + comment_line_str ="#" ; + }else if (value[0 ]) {if (strchr (value,'\n' ))return error (_ ("%s cannot contain newline" ), var); comment_line_str = value;
This patch is exactly what Phillip suggested in
https://lore.kernel.org/git/[email protected]/
isn't it? Makes sense to me.
Re: [GSOC PATCH 2/2] config: set comment_line_str to "#" when core.commentChar=auto
Hi Junio,
Quoted 1 line · Junio C Hamano
[snip]
Quoted 7 lines
This patch is exactly what Phillip suggested in
https://lore.kernel.org/git/[email protected]/
isn't it? Makes sense to me.
Yes, you're right. I should add the suggested-by trailer for this patch.
Thanks
Ayush
Re: [GSOC PATCH 2/2] config: set comment_line_str to "#" when core.commentChar=auto
Quoted 4 lines · Ayush Chandekar
Quoted 1 line · Junio C Hamano
[snip]
Quoted 1 line
Quoted 8 lines
Quoted 6 lines
This patch is exactly what Phillip suggested in
https://lore.kernel.org/git/[email protected]/
isn't it? Makes sense to me.
Yes, you're right. I should add the suggested-by trailer for this patch.
I am not sure about that, though. A verbatim copy is stronger than implementing what was suggested by another person. If I were in your position, I'll probably say something like
The patch text was taken from Phillip Wood's message [*URL*], with the commit log message written by me.
Based-on-a-patch-by: Phillip Wood <...>
Signed-off-by: Ayush Chandekar <...>
In any case, this overlaps both textually but also intent-wise with Phillip's "let's mark core.commentchar=auto deprecated and remove the support at 3.0 boundary", which is planned to be rerolled to make it a failure when the user uses core.commentchar=auto. It would be a while before we tag Git 3.0, so the fix in this topic will be necessary until then.
Thanks.
Re: [GSOC PATCH 2/2] config: set comment_line_str to "#" when core.commentChar=auto
Quoted 26 lines · Junio C Hamano
Quoted 4 lines · Ayush Chandekar
Quoted 1 line · Junio C Hamano
[snip]
Quoted 1 line
Quoted 8 lines
Quoted 6 lines
This patch is exactly what Phillip suggested in
https://lore.kernel.org/git/[email protected]/
isn't it? Makes sense to me.
Yes, you're right. I should add the suggested-by trailer for this patch.
I am not sure about that, though. A verbatim copy is stronger than implementing what was suggested by another person. If I were in your position, I'll probably say something like
The patch text was taken from Phillip Wood's message [*URL*], with the commit log message written by me.
Based-on-a-patch-by: Phillip Wood <...>
Signed-off-by: Ayush Chandekar <...>
Quoted 9 lines
In any case, this overlaps both textually but also intent-wise with Phillip's "let's mark core.commentchar=auto deprecated and remove the support at 3.0 boundary", which is planned to be rerolled to make it a failure when the user uses core.commentchar=auto. It would be a while before we tag Git 3.0, so the fix in this topic will be necessary until then.
Thanks.
Yeah, Phillip should actually get the primary credit for this patch and Suggested-by does not do enough justice.
I will send a new version right away.
Thanks!
Ayush
Re: [GSOC PATCH 2/2] config: set comment_line_str to "#" when core.commentChar=auto
Quoted 3 lines · Ayush Chandekar
Yeah, Phillip should actually get the primary credit for this patch and Suggested-by does not do enough justice.
I will send a new version right away.
Thanks. Don't forget to ask him to sign-off.
Re: [GSOC PATCH 2/2] config: set comment_line_str to "#" when core.commentChar=auto
Quoted 7 lines · Junio C Hamano
Quoted 3 lines · Ayush Chandekar
Yeah, Phillip should actually get the primary credit for this patch and Suggested-by does not do enough justice.
I will send a new version right away.Thanks. Don't forget to ask him to sign-off.
Here it is
Re: [GSOC PATCH 2/2] config: set comment_line_str to "#" when core.commentChar=auto
Quoted 11 lines · Phillip Wood
Quoted 6 lines · Junio C Hamano
Quoted 3 lines · Ayush Chandekar
Yeah, Phillip should actually get the primary credit for this patch and Suggested-by does not do enough justice.
I will send a new version right away.Thanks. Don't forget to ask him to sign-off.
Here it is
Signed-off-by <Phillip Wood [email protected]>
Thanks.
v5 cover commit: improve behaviour of core.commentChar=auto for comments in commit messages
In reply to Ayush ChandekarHey everyone,
The aim of this patch series is to improve the behaviour of core.commentChar=auto by the following patches:
1/2 - Fix a bug which reads comment character of the comments in commit message leading to change in the value of comment_line_str and thus resulting the comments in the final commit message. 2/2 - Standardizes the behaviour of code by resetting the 'comment_line_str' to "#" when 'core.commentChar' is set to "auto".
Thanks to Junio, Phillip and Kristoffer for reviewing the patches and also Christian for the reviews and mentoring me.
The only difference between this version (v5) and the previous one is that I've added credit to Phillip for patch (2/2).
Ayush Chandekar (2):
commit: avoid scanning trailing comments when 'core.commentChar' is
"auto"
config: set comment_line_str to "#" when core.commentChar=auto
builtin/commit.c | 6 +++++- config.c | 6 ++++-- t/t3418-rebase-continue.sh | 13 +++++++++++++ 3 files changed, 22 insertions(+), 3 deletions(-)
Signature · 1 line
2.49.0
PATCH v5 1/2 commit: avoid scanning trailing comments when 'core.commentChar' is "auto"
When core.commentChar is set to "auto", Git selects a comment character by scanning the commit message contents and avoiding any character already present in the message.
If the message still contains old conflict comments (starting with a comment character), Git assumes that character is in use and chooses a different one. As a result, those existing comment lines are no longer recognized as comments and end up being included in the final commit message.
To avoid this, skip scanning the trailing comment block when selecting the comment character. This allows Git to safely reuse the original character when appropriate, keeping the commit message clean and free of leftover conflict information.
Background:
The "auto" value for core.commentchar was introduced in the commit 84c9dc2c5a (commit: allow core.commentChar=auto for character auto selection, 2014-05-17) but did not exhibit this issue at that time.
The bug was introduced in commit a6c2654f83 (rebase -m: fix --signoff with conflicts, 2024-04-18) where Git started writing conflict comments to the file at 'rebase_path_message()'.
--- builtin/commit.c | 6 +++++- t/t3418-rebase-continue.sh | 13 +++++++++++++ 2 files changed, 18 insertions(+), 1 deletion(-)
builtin/commit.c +5 -1
@@ -688,6 +688,10 @@ static void adjust_comment_line_char(const struct strbuf *sb)char candidates[] ="#;@!$%^&|:" ;char *candidate;const char *p; +size_t cutoff; + +/* Ignore comment chars in trailing comments (e.g., Conflicts:) */ + cutoff = sb->len -ignored_log_message_bytes (sb->buf, sb->len);if (!memchr (sb->buf, candidates[0 ], sb->len)) {free (comment_line_str_to_free);
@@ -700,7 +704,7 @@ static void adjust_comment_line_char(const struct strbuf *sb) candidate =strchr (candidates, *p);if (candidate) *candidate =' ' ; -for (p = sb->buf; *p; p++) { +for (p = sb->buf; p +1 < sb->buf + cutoff; p++) {if ((p[0 ] =='\n' || p[0 ] =='\r' ) && p[1 ]) { candidate =strchr (candidates, p[1 ]);if (candidate)
t/t3418-rebase-continue.sh +13 -0
@@ -328,6 +328,19 @@ test_expect_success 'there is no --no-reschedule-failed-exec in an ongoing rebas test_expect_code 129 git rebase --edit-todo --no-reschedule-failed-exec ' +test_expect_success 'no change in comment character due to conflicts markers with core.commentChar=auto' ' + git checkout -b branch-a && + test_commit A F1 && + git checkout -b branch-b HEAD^ && + test_commit B F1 && + test_must_fail git rebase branch-a && + printf "B\nA\n" >F1 && + git add F1 && + GIT_EDITOR="cat >actual" git -c core.commentChar=auto rebase --continue && + # Check that "#" is still the comment character. + test_grep "^# Changes to be committed" actual +' + test_orig_head_helper () { test_when_finished 'git rebase --abort && git checkout topic &&
Signature · 1 line
2.49.0
PATCH v5 2/2 config: set comment_line_str to "#" when core.commentChar=auto
In reply to Ayush ChandekarIf conflict comments already use a comment character that isn't "#", and core.commentChar is set "auto", Git will ignore these lines during the scan using ignored_log_message_bytes() and pick a new comment character based on the rest of the message. The newly chosen character may be different from the one used in the conflict comments and therefore, these are no longer treated as comments and end up in the final commit message.
For example, during a rebase if the user previously set core.commentChar=% and then encounters a conflict, conflict comments like "% Conflicts:" are generated. If the user subsequently sets core.commentChar=auto before running rebase --continue, Git parses the "auto" setting and begins scanning. It first uses the existing 'comment_line_str' (which is '%') to detect and ignore conflict comments via ignored_log_message_bytes().
Then, Git scans the rest of the message (excluding conflict comments), sees that none of the remaining lines start with '#' and decides to set comment_line_str to '#'. Since the final commit character differs from the one used in the conflict comments, those lines are no longer considered comments and get included in the final commit message.
Set 'comment_line_str' to '#' when core.commentChar is set to 'auto' to reset any previously set value.
While this does not solve the issue of conflict comment inclusion and the user visible behaviour stays tha same, it standardizes the behaviour of the code by always resetting 'comment_line_str' to '#' when core.commentChar=auto is parsed.
The patch text is based on Phillip Wood's message:
https://lore.kernel.org/git/[email protected]/
and the commit log message is wriiten by me.
--- config.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
config.c +4 -2
@@ -1537,9 +1537,11 @@ static int git_default_core_config(const char *var, const char *value, !strcmp(var, "core.commentstring")) {if (!value)return config_error_nonbool (var); -else if (!strcasecmp (value,"auto" )) +else if (!strcasecmp (value,"auto" )) { auto_comment_line_char =1 ; -else if (value[0 ]) { +FREE_AND_NULL (comment_line_str_to_free); + comment_line_str ="#" ; + }else if (value[0 ]) {if (strchr (value,'\n' ))return error (_ ("%s cannot contain newline" ), var); comment_line_str = value;
Signature · 1 line
2.49.0
Re: [GSOC PATCH v5 2/2] config: set comment_line_str to "#" when core.commentChar=auto
Quoted 11 lines · Ayush Chandekar
If conflict comments already use a comment character that isn't "#", and
...
The patch text is based on Phillip Wood's message:
https://lore.kernel.org/git/[email protected]/
and the commit log message is wriiten by me.Based-on-a-patch-by Phillip Wood <[email protected]>Mentored-by Christian Couder <[email protected]>Mentored-by Ghanshyam Thakkar <[email protected]>Signed-off-by Ayush Chandekar <[email protected]>---
Earlier in response to your "Phillip should actually get the primary credit" I said to ask for his sign-off, because I took it as you are actually making Phillip the author of the patch. But it is fine either way. Phillip has given his permission to add a sign-off, so we have everything to move this topic forward.
Thanks, all!
Quoted 21 lines
config.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)config.c +4
-2@@ -1537,9 +1537,11 @@ static int git_default_core_config(const char *var, const char *value, !strcmp(var, "core.commentstring")) {if (!value)return config_error_nonbool (var); -else if (!strcasecmp (value,"auto" )) +else if (!strcasecmp (value,"auto" )) { auto_comment_line_char =1 ; -else if (value[0 ]) { +FREE_AND_NULL (comment_line_str_to_free); + comment_line_str ="#" ; + }else if (value[0 ]) {if (strchr (value,'\n' ))return error (_ ("%s cannot contain newline" ), var); comment_line_str = value;
Re: [GSOC PATCH v5 0/2] commit: improve behaviour of core.commentChar=auto for comments in commit messages
In reply to Ayush ChandekarHi Ayush
Quoted 6 lines · Ayush Chandekar
Hey everyone,
The aim of this patch series is to improve the behaviour of core.commentChar=auto by the following patches:
1/2 - Fix a bug which reads comment character of the comments in commit message leading to change in the value ofcomment_line_strand thus resulting the comments in the final commit message. 2/2 - Standardizes the behaviour of code by resetting the 'comment_line_str' to "#" when 'core.commentChar' is set to "auto".
This version looks good to me, thanks for working on it.
Junio - shall I rebase 'pw/3.0-commentchar-auto-deprecation' on top of this when I re-roll to avoid conflicts?
Thanks
Phillip
Quoted 6 lines
Thanks to Junio, Phillip and Kristoffer for reviewing the patches and also Christian for the reviews and mentoring me.
The only difference between this version (v5) and the previous one is that I've added credit to Phillip for patch (2/2).
Ayush Chandekar (2):
commit: avoid scanning trailing comments when 'core.commentChar' is
Quoted 8 lines
"auto"config: set
comment_line_strto "#" when core.commentChar=autobuiltin/commit.c | 6 +++++- config.c | 6 ++++-- t/t3418-rebase-continue.sh | 13 +++++++++++++ 3 files changed, 22 insertions(+), 3 deletions(-)
J
Re: [GSOC PATCH v5 0/2] commit: improve behaviour of core.commentChar=auto for comments in commit messages
Quoted 4 lines · Phillip Wood
This version looks good to me, thanks for working on it.
Junio - shall I rebase 'pw/3.0-commentchar-auto-deprecation' on top of this when I re-roll to avoid conflicts?
Sounds sensible. I can drop my merge-fix then.
Thanks.