Remove unused var
PATCH 1/3 Remove unused var
--- t/t4060-diff-submodule-option-diff-format.sh | 1 - 1 file changed, 1 deletion(-)
t/t4060-diff-submodule-option-diff-format.sh +0 -1
@@ -361,7 +361,6 @@ test_expect_success 'typechanged submodule(submodule->blob)' ' rm -f sm1 && test_create_repo sm1 && head6=$(add_file sm1 foo6 foo7) -fullhead6=$(cd sm1; git rev-parse --verify HEAD) test_expect_success 'nonexistent commit' ' git diff-index -p --submodule=diff HEAD >actual && cat >expected <<-EOF &&
Signature · 1 line
2.11.GIT
PATCH 2/3 diff --submodule=diff: do not fail on ever-initialied deleted submodules
If you have ever initialized a submodule, open_submodule will open it. If you then delete the submodule's worktree directory (but don't remove it from .gitmodules), git diff --submodule=diff would crash as it attempted to chdir into the now-deleted working tree directory.
--- submodule.c | 3 ++ t/t4060-diff-submodule-option-diff-format.sh | 45 ++++++++++++++++++++++++---- 2 files changed, 43 insertions(+), 5 deletions(-)
submodule.c +3 -0
@@ -673,6 +673,9 @@ void show_submodule_inline_diff(struct diff_options *o, const char *path, !(right || is_null_oid(two))) goto done; +if (!is_directory (path)) + goto done; +if (left) old_oid = one;if (right)
t/t4060-diff-submodule-option-diff-format.sh +40 -5
71 lines not on this page · raw message
Signature · 1 line
2.11.GIT
Re: [PATCH 2/3] diff --submodule=diff: do not fail on ever-initialied deleted submodules
Quoted 4 lines · David Turner
If you have ever initialized a submodule,
open_submodulewill open it. If you then delete the submodule's worktree directory (but don't remove it from .gitmodules), git diff --submodule=diff would crash as it attempted to chdir into the now-deleted working tree directory.
Hmph. So what does the failure look like? The child process inside start_command() attempts chdir() and reports CHILD_ERR_CHDIR back, and we catch it as an error by reading from notify_pipe[0] and report failure from start_command()? Or are we talking about a more severe "crash" of an uncontrolled kind?
Bypassing the execution of diff in the submodule like this patch does may avoid such a failure, but is that all we need to "fix" this issue? What does the user expect after removing a submodule that way and runs "diff" with the "--submodule=diff" option? Shouldn't we be giving "all lines from all files have been removed" patch?
Thanks.
Quoted 98 lines
Signed-off-by David Turner <[email protected]>--- submodule.c | 3 ++ t/t4060-diff-submodule-option-diff-format.sh | 45 ++++++++++++++++++++++++---- 2 files changed, 43 insertions(+), 5 deletions(-)submodule.c +3
-0@@ -673,6 +673,9 @@ void show_submodule_inline_diff(struct diff_options *o, const char *path, !(right || is_null_oid(two))) goto done; +if (!is_directory (path)) + goto done; +if (left) old_oid = one;if (right)t/t4060-diff-submodule-option-diff-format.sh +40
-5@@ -703,10 +703,26 @@ test_expect_success 'path filter' ' diff_cmp expected actual ' -commit_file sm2 +cat >.gitmodules <<-EOF +[submodule "sm2"] + path = sm2 + url = bogus_url +EOF +git add .gitmodules +commit_file sm2 .gitmodules + test_expect_success 'given commit' ' git diff-index -p --submodule=diff HEAD^ >actual && cat >expected <<-EOF && + diff --git a/.gitmodules b/.gitmodules + new file mode 100644 + index 1234567..89abcde + --- /dev/null + +++ b/.gitmodules + @@ -0,0 +1,3 @@ + +[submodule "sm2"] + +path = sm2 + +url = bogus_url Submodule sm1 $head7...0000000 (submodule deleted) Submodule sm2 0000000...$head9 (new submodule) diff --git a/sm2/foo8 b/sm2/foo8@@ -728,15 +744,21 @@ test_expect_success 'given commit' ' ' test_expect_success 'setup .git file for sm2' ' - (cd sm2 && - REAL="$(pwd)/../.real" && - mv .git "$REAL" && - echo "gitdir: $REAL" >.git) + git submodule absorbgitdirs sm2 ' test_expect_success 'diff --submodule=diff with .git file' ' git diff --submodule=diff HEAD^ >actual && cat >expected <<-EOF && + diff --git a/.gitmodules b/.gitmodules + new file mode 100644 + index 1234567..89abcde + --- /dev/null + +++ b/.gitmodules + @@ -0,0 +1,3 @@ + +[submodule "sm2"] + +path = sm2 + +url = bogus_url Submodule sm1 $head7...0000000 (submodule deleted) Submodule sm2 0000000...$head9 (new submodule) diff --git a/sm2/foo8 b/sm2/foo8@@ -757,6 +779,19 @@ test_expect_success 'diff --submodule=diff with .git file' ' diff_cmp expected actual ' +mv sm2 sm2-bak + +test_expect_success 'deleted submodule with .git file' ' + git diff-index -p --submodule=diff HEAD >actual && + cat >expected <<-EOF && + Submodule sm1 $head7...0000000 (submodule deleted) + Submodule sm2 $head9...0000000 (submodule deleted) + EOF + diff_cmp expected actual +' + +mv sm2-bak sm2 + test_expect_success 'setup nested submodule' ' git submodule add -f ./sm2 && git commit -a -m "add sm2" &&
RE: [PATCH 2/3] diff --submodule=diff: do not fail on ever-initialied deleted submodules
Quoted 14 lines
From: Junio C Hamano <[email protected]>
Sent: Monday, July 26, 2021 6:57 PM
To: David Turner <[email protected]>
Cc: [email protected]
Subject: Re: [PATCH 2/3] diff --submodule=diff: do not fail on ever-initialied deleted submodulesQuoted 4 lines · David Turner
If you have ever initialized a submodule,
open_submodulewill open it. If you then delete the submodule's worktree directory (but don't remove it from .gitmodules), git diff --submodule=diff would crash as it attempted to chdir into the now-deleted working tree directory.
Quoted 5 lines
Hmph. So what does the failure look like? The child process inside
start_command()attemptschdir()and reportsCHILD_ERR_CHDIRback, and we catch it as an error by reading from notify_pipe[0] and report failure fromstart_command()? Or are we talking about a more severe "crash" of an uncontrolled kind?
It's the first kind.
Quoted 5 lines
Bypassing the execution of diff in the submodule like this patch does may avoid such a failure, but is that all we need to "fix" this issue? What does the user expect after removing a submodule that way and runs "diff" with the "-- submodule=diff" option? Shouldn't we be giving "all lines from all files have been removed" patch?
Just a note: this only matters if the submodules git dir is absorbed. If not, then we no longer have anywhere to run the diff. But that case does not trigger this error, because in that case, open_submodule fails, so we don't resolve a left commit, so we exit early, which is the only thing we could do.
If absorbed, then we could, in theory, go into the submodule's absorbed git dir (.git/modules/sm2) and run the diff there. But in practice, that's a bit more complicated, because git diff expects to be run from inside a working directory, not a git dir. So it looks in the config for core.worktree, and does chdir("../../../sm2"), which is the very dir that we're trying to avoid visiting because it's been deleted. We could work around this by setting GIT_WORK_TREE (and GIT_DIR) to ".". This actually seems to work, but it's a little weird to set GIT_WORK_TREE to something that is not a working tree just to avoid an unnecessary chdir. To my mild surprise, it also seems to work correctly in the case of deleted nested (absorbed) submodules. What do you think of this idea?
(Side note: The same bit about chdir into the working tree is true for diff-tree even though it normally does not need anything from the working tree. I say "normally", because in the case of nested submodules, it might need to look inside those submodules, which might themselves not be absorbed. Of course, this case cannot obtain if the submodule in the worktree has been deleted. We should consider fixing the docs for git diff-tree --submodule=diff to specify that it only works if -p is passed.)
(Sorry if the formatting on this email ends up bad -- corporate email is... corporate . I'm going to CC my personal address so that I can use a better mailer on future replies).
Re: [PATCH 2/3] diff --submodule=diff: do not fail on ever-initialied deleted submodules
Quoted 57 lines · David Turner
Quoted 14 lines
From: Junio C Hamano <[email protected]>
Sent: Monday, July 26, 2021 6:57 PM
To: David Turner <[email protected]>
Cc: [email protected]
Subject: Re: [PATCH 2/3] diff --submodule=diff: do not fail on ever-initialied deleted submodulesQuoted 4 lines · David Turner
If you have ever initialized a submodule,
open_submodulewill open it. If you then delete the submodule's worktree directory (but don't remove it from .gitmodules), git diff --submodule=diff would crash as it attempted to chdir into the now-deleted working tree directory.Quoted 5 lines
Hmph. So what does the failure look like? The child process inside
start_command()attemptschdir()and reportsCHILD_ERR_CHDIRback, and we catch it as an error by reading from notify_pipe[0] and report failure fromstart_command()? Or are we talking about a more severe "crash" of an uncontrolled kind?It's the first kind.
Quoted 5 lines
Bypassing the execution of diff in the submodule like this patch does may avoid such a failure, but is that all we need to "fix" this issue? What does the user expect after removing a submodule that way and runs "diff" with the "-- submodule=diff" option? Shouldn't we be giving "all lines from all files have been removed" patch?
Just a note: this only matters if the submodules git dir is absorbed. If not, then we no longer have anywhere to run the diff. But that case does not trigger this error, because in that case,
open_submodulefails, so we don't resolve a left commit, so we exit early, which is the only thing we could do.If absorbed, then we could, in theory, go into the submodule's absorbed git dir (.git/modules/sm2) and run the diff there. But in practice, that's a bit more complicated, because
git diffexpects to be run from inside a working directory, not a git dir. So it looks in the config for core.worktree, and doeschdir("../../../sm2"), which is the very dir that we're trying to avoid visiting because it's been deleted. We could work around this by settingGIT_WORK_TREE(andGIT_DIR) to ".". This actually seems to work, but it's a little weird to setGIT_WORK_TREEto something that is not a working tree just to avoid an unnecessary chdir. To my mild surprise, it also seems to work correctly in the case of deleted nested (absorbed) submodules. What do you think of this idea?(Side note: The same bit about chdir into the working tree is true for diff-tree even though it normally does not need anything from the working tree. I say "normally", because in the case of nested submodules, it might need to look inside those submodules, which might themselves not be absorbed. Of course, this case cannot obtain if the submodule in the worktree has been deleted. We should consider fixing the docs for git diff-tree --submodule=diff to specify that it only works if -p is passed.)
Quoted 3 lines
(Sorry if the formatting on this email ends up bad -- corporate email is... corporate . I'm going to CC my personal address so that I can use a better mailer on future replies).
That I had to ask questions based on the proposed log message and you needed to answer to clarify means that the next person who encounters this commit in "git log" would likely have to ask the same question, and worse, unlike I had you, there is nobody to whom they ask for help understanding this commit.
Thanks.
PATCH v4 1/3 Remove unused var
--- t/t4060-diff-submodule-option-diff-format.sh | 1 - 1 file changed, 1 deletion(-)
t/t4060-diff-submodule-option-diff-format.sh +0 -1
@@ -361,7 +361,6 @@ test_expect_success 'typechanged submodule(submodule->blob)' ' rm -f sm1 && test_create_repo sm1 && head6=$(add_file sm1 foo6 foo7) -fullhead6=$(cd sm1; git rev-parse --verify HEAD) test_expect_success 'nonexistent commit' ' git diff-index -p --submodule=diff HEAD >actual && cat >expected <<-EOF &&
Signature · 1 line
2.11.GIT
PATCH v4 2/3 diff --submodule=diff: do not fail on ever-initialied deleted submodules
If you have ever initialized a submodule, open_submodule will open it. If you then delete the submodule's worktree directory (but don't remove it from .gitmodules), git diff --submodule=diff would error out as it attempted to chdir into the now-deleted working tree directory.
This only matters if the submodules git dir is absorbed. If not, then we no longer have anywhere to run the diff. But that case does not trigger this error, because in that case, open_submodule fails, so we don't resolve a left commit, so we exit early, which is the only thing we could do.
If absorbed, then we can run the diff from the submodule's absorbed git dir (.git/modules/sm2). In practice, that's a bit more complicated, because git diff expects to be run from inside a working directory, not a git dir. So it looks in the config for core.worktree, and does chdir("../../../sm2"), which is the very dir that we're trying to avoid visiting because it's been deleted. We work around this by setting GIT_WORK_TREE (and GIT_DIR) to ".". It is little weird to set GIT_WORK_TREE to something that is not a working tree just to avoid an unnecessary chdir, but it works.
--- submodule.c | 10 ++ t/t4060-diff-submodule-option-diff-format.sh | 158 +++++++++++++++++++++++++-- 2 files changed, 161 insertions(+), 7 deletions(-)
submodule.c +10 -0
@@ -710,6 +710,16 @@ void show_submodule_inline_diff(struct diff_options *o, const char *path,strvec_push (&cp.args,oid_to_hex (new_oid));prepare_submodule_repo_env (&cp.env_array); + +if (!is_directory (path)) { +/* fall back to absorbed git dir, if any */ +if (!sub) + goto done; + cp.dir = sub->gitdir; +strvec_push (&cp.env_array,GIT_DIR_ENVIRONMENT "=." ); +strvec_push (&cp.env_array,GIT_WORK_TREE_ENVIRONMENT "=." ); + } +if (start_command (&cp))diff_emit_submodule_error (o,"(diff failed)\n" );
t/t4060-diff-submodule-option-diff-format.sh +151 -7
195 lines not on this page · raw message
Signature · 1 line
2.11.GIT
PATCH v4 3/3 diff --submodule=diff: Don't print failure message twice
In reply to David TurnerWhen we fail to start a diff command inside a submodule, immediately exit the routine rather than trying to finish the command and printing a second message.
--- submodule.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
submodule.c +3 -1
@@ -720,8 +720,10 @@ void show_submodule_inline_diff(struct diff_options *o, const char *path,strvec_push (&cp.env_array,GIT_WORK_TREE_ENVIRONMENT "=." ); } -if (start_command (&cp)) +if (start_command (&cp)) {diff_emit_submodule_error (o,"(diff failed)\n" ); + goto done; + }while (strbuf_getwholeline_fd (&sb, cp.out,'\n' ) !=EOF )diff_emit_submodule_pipethrough (o, sb.buf, sb.len);
Signature · 1 line
2.11.GIT
Quoted 4 lines · David Turner
Signed-off-by David Turner <[email protected]>--- t/t4060-diff-submodule-option-diff-format.sh | 1 - 1 file changed, 1 deletion(-)
Of course this still makes sense ;-)
Let me retitle it to "t4060: remove unused variable" to help readers of "git shortlog --no-merges", though, while queuing.
Other two patches looked sensible to me, but I'd welcome comments by submodule users, of course ;-)
Thanks.
Quoted 12 lines
t/t4060-diff-submodule-option-diff-format.sh +0
-1@@ -361,7 +361,6 @@ test_expect_success 'typechanged submodule(submodule->blob)' ' rm -f sm1 && test_create_repo sm1 && head6=$(add_file sm1 foo6 foo7) -fullhead6=$(cd sm1; git rev-parse --verify HEAD) test_expect_success 'nonexistent commit' ' git diff-index -p --submodule=diff HEAD >actual && cat >expected <<-EOF &&
PATCH 3/3 diff --submodule=diff: Don't print failure message twice
In reply to David TurnerWhen we fail to start a diff command inside a submodule, immediately exit the routine rather than trying to finish the command and printing a second message.
--- submodule.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
submodule.c +3 -1
@@ -713,8 +713,10 @@ void show_submodule_inline_diff(struct diff_options *o, const char *path,strvec_push (&cp.args,oid_to_hex (new_oid));prepare_submodule_repo_env (&cp.env_array); -if (start_command (&cp)) +if (start_command (&cp)) {diff_emit_submodule_error (o,"(diff failed)\n" ); + goto done; + }while (strbuf_getwholeline_fd (&sb, cp.out,'\n' ) !=EOF )diff_emit_submodule_pipethrough (o, sb.buf, sb.len);
Signature · 1 line
2.11.GIT
Re: [PATCH 3/3] diff --submodule=diff: Don't print failure message twice
Quoted 15 lines · David Turner
When we fail to start a diff command inside a submodule, immediately exit the routine rather than trying to finish the command and printing a second message.
Signed-off-by David Turner <[email protected]>--- submodule.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)submodule.c +0
-0@@ -713,8 +713,10 @@ void show_submodule_inline_diff(struct diff_options *o, const char *path,strvec_push (&cp.args,oid_to_hex (new_oid));
submodule.c void show_submodule_inline_diff(struct diff_options *o, const char *path, · line 722
Quoted 10 lines
submodule.c +3
-1prepare_submodule_repo_env (&cp.env_array); -if (start_command (&cp)) +if (start_command (&cp)) {diff_emit_submodule_error (o,"(diff failed)\n" ); + goto done; + }while (strbuf_getwholeline_fd (&sb, cp.out,'\n' ) !=EOF )diff_emit_submodule_pipethrough (o, sb.buf, sb.len);
Again, makes sense. Thanks.
Quoted 10 lines · David Turner
Signed-off-by David Turner <[email protected]>--- t/t4060-diff-submodule-option-diff-format.sh | 1 - 1 file changed, 1 deletion(-)t/t4060-diff-submodule-option-diff-format.sh +0
-0
t/t4060-diff-submodule-option-diff-format.sh test_expect_success 'typechanged submodule(submodule->blob)' ' · line 366
Quoted 7 lines
t/t4060-diff-submodule-option-diff-format.sh +0
-1@@ -361,7 +361,6 @@ test_expect_success 'typechanged submodule(submodule->blob)' ' rm -f sm1 && test_create_repo sm1 && head6=$(add_file sm1 foo6 foo7) -fullhead6=$(cd sm1; git rev-parse --verify HEAD) test_expect_success 'nonexistent commit' ' git diff-index -p --submodule=diff HEAD >actual && cat >expected <<-EOF &&
Makes sense. THanks.