threads / discuss / 51490

Merge: "git rm bla": "bla: needs merge", but still removes file "bla"

Subject: Merge: "git rm bla": "bla: needs merge", but still removes file "bla"

## tl;dr

6 messages between Jul 17, 2019 and Jul 18, 2019.

replies: 5people: 3as markdown or json

Ulrich Windl· Jul 17, 2019, 07:58 UTC · lore
Hi!

I just had "an interesting case" for a merge with conflicts: The merge re-introduced a file that had been renamed (say old one is "bla", and the new one is "foo"). After merging the changes from bla into foo, I added foo, trying to remove bla:

> git add foo
> git rm bla

bla: needs merge rm 'bla'

The interesting part is that the file was still removed, while claiming "needs merge" before:
> git rm bla
fatal: pathspec 'bla' did not match any files
(git-2.12.3-27.14.1.x86_64 from SLES12 SP4)

Regards, Ulrich

Junio C Hamano· Jul 17, 2019, 16:44 UTC · re: Ulrich Windl · lore

Re: Merge: "git rm bla": "bla: needs merge", but still removes file "bla"

"Ulrich Windl" <Ulrich.Windl@rz.uni-regensburg.de> writes:
Show 7 quoted lines
> I just had "an interesting case" for a merge with conflicts:
> The merge re-introduced a file that had been renamed (say old one is "bla", and the new one is "foo").
> After merging the changes from bla into foo, I added foo, trying to remove bla:
>> git add foo
>> git rm bla
> bla: needs merge
> rm 'bla'

Yeah, I think I've known about this for a long time. We have an internal call to "update-index --refresh" before starting to remove, because we need to know which path is up-to-date wrt the index, and the machinery for refreshing the index by default gives the "needs merge" message. We never bothered to squelch it.

Perhaps this one-liner would be a sufficient fix.
 builtin/rm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/builtin/rm.c b/builtin/rm.c
index 2eacda42b4..19ce95a901 100644
--- a/builtin/rm.c
+++ b/builtin/rm.c
@@ -273,7 +273,7 @@ int cmd_rm(int argc, const char **argv, const char *prefix)
 	parse_pathspec(&pathspec, 0,
 		       PATHSPEC_PREFER_CWD,
 		       prefix, argv);
-	refresh_index(&the_index, REFRESH_QUIET, &pathspec, NULL, NULL);
+	refresh_index(&the_index, REFRESH_QUIET|REFRESH_UNMERGED, &pathspec, NULL, NULL);
 
 	seen = xcalloc(pathspec.nr, 1);
Junio C Hamano· Jul 17, 2019, 20:38 UTC · re: Junio C Hamano · lore

[PATCH] rm: resolving by removal is not a warning-worthy event

When resolving a conflict on a path in favor of removing it, using "git rm" on it is the standard way to do so. The user however is greeted with a "needs merge" message during that operation:

	$ git merge side-branch
	$ edit conflicted-path-1
	$ git add conflicted-path-1
	$ git rm conflicted-path-2
	conflicted-path-2: needs merge
	rm 'conflicted-path-2'

The removal by "git rm" does get performed, but an uninitiated user may find it confusing, "needs merge? so I need to resolve conflict before being able to remove it???"

The message is coming from "update-index --refresh" that is called internally to make sure "git rm" knows which paths are clean and which paths are dirty, in order to prevent removal of paths modified relative to the index without the "-f" option. We somehow ended up not squelching this message which seeped through to the UI surface.

Use the same mechanism used by "git commit", "git describe", etc. to squelch the message.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
---
 builtin/rm.c  |  2 +-
 t/t3600-rm.sh | 13 +++++++++++++
 2 files changed, 14 insertions(+), 1 deletion(-)
diff --git a/builtin/rm.c b/builtin/rm.c
index 65b448ef8e..b63c86ae92 100644
--- a/builtin/rm.c
+++ b/builtin/rm.c
@@ -272,7 +272,7 @@ int cmd_rm(int argc, const char **argv, const char *prefix)
 	parse_pathspec(&pathspec, 0,
 		       PATHSPEC_PREFER_CWD,
 		       prefix, argv);
-	refresh_index(&the_index, REFRESH_QUIET, &pathspec, NULL, NULL);
+	refresh_index(&the_index, REFRESH_QUIET|REFRESH_UNMERGED, &pathspec, NULL, NULL);
 
 	seen = xcalloc(pathspec.nr, 1);
 
diff --git a/t/t3600-rm.sh b/t/t3600-rm.sh
index b8fbdefcdc..5aae78ccc4 100755
--- a/t/t3600-rm.sh
+++ b/t/t3600-rm.sh
@@ -251,6 +251,19 @@ test_expect_success 'choking "git rm" should not let it die with cruft' '
 	test_path_is_missing .git/index.lock
 '
 
+test_expect_success 'Resolving by removal is not a warning-worthy event' '
+	git reset -q --hard &&
+	test_when_finished "rm -f .git/index.lock msg && git reset -q --hard" &&
+	qfwfq=$(echo qfwfq | git hash-object -w --stdin) &&
+	for stage in 1 2 3
+	do
+		echo "100644 $qfwfq $stage	qfwfq"
+	done | git update-index --index-info &&
+	git rm qfwfq >msg &&
+	test_i18ngrep ! "needs merge" msg &&
+	test_must_fail git ls-files -s --error-unmatch qfwfq
+'
+
 test_expect_success 'rm removes subdirectories recursively' '
 	mkdir -p dir/subdir/subsubdir &&
 	echo content >dir/subdir/subsubdir/file &&

-- 
2.22.0-653-g37fc7794bc
Jeff King· Jul 18, 2019, 20:26 UTC · re: Junio C Hamano · lore

Re: [PATCH] rm: resolving by removal is not a warning-worthy event

On Wed, Jul 17, 2019 at 01:38:35PM -0700, Junio C Hamano wrote:
Show 23 quoted lines
> When resolving a conflict on a path in favor of removing it, using
> "git rm" on it is the standard way to do so.  The user however is
> greeted with a "needs merge" message during that operation:
> 
> 	$ git merge side-branch
> 	$ edit conflicted-path-1
> 	$ git add conflicted-path-1
> 	$ git rm conflicted-path-2
> 	conflicted-path-2: needs merge
> 	rm 'conflicted-path-2'
> 
> The removal by "git rm" does get performed, but an uninitiated user
> may find it confusing, "needs merge? so I need to resolve conflict
> before being able to remove it???"
> 
> The message is coming from "update-index --refresh" that is called
> internally to make sure "git rm" knows which paths are clean and
> which paths are dirty, in order to prevent removal of paths modified
> relative to the index without the "-f" option.  We somehow ended up
> not squelching this message which seeped through to the UI surface.
> 
> Use the same mechanism used by "git commit", "git describe", etc. to
> squelch the message.
Nicely explained, and the patch makes sense.
> +test_expect_success 'Resolving by removal is not a warning-worthy event' '
> +	git reset -q --hard &&
> +	test_when_finished "rm -f .git/index.lock msg && git reset -q --hard" &&
> +	qfwfq=$(echo qfwfq | git hash-object -w --stdin) &&

I'd have called this "$blob" for my own sanity in typing later lines, but OK. :)

Show 5 quoted lines
> +	do
> +		echo "100644 $qfwfq $stage	qfwfq"
> +	done | git update-index --index-info &&
> +	git rm qfwfq >msg &&
> +	test_i18ngrep ! "needs merge" msg &&
Should we capture stderr from "git rm", too, to cover all bases?
-Peff
Junio C Hamano· Jul 18, 2019, 21:07 UTC · re: Jeff King · lore

Re: [PATCH] rm: resolving by removal is not a warning-worthy event

Jeff King <peff@peff.net> writes:
Show 7 quoted lines
>> +test_expect_success 'Resolving by removal is not a warning-worthy event' '
>> +	git reset -q --hard &&
>> +	test_when_finished "rm -f .git/index.lock msg && git reset -q --hard" &&
>> +	qfwfq=$(echo qfwfq | git hash-object -w --stdin) &&
>
> I'd have called this "$blob" for my own sanity in typing later lines,
> but OK. :)
OK, I can change that easily ;-)
Show 7 quoted lines
>> +	do
>> +		echo "100644 $qfwfq $stage	qfwfq"
>> +	done | git update-index --index-info &&
>> +	git rm qfwfq >msg &&
>> +	test_i18ngrep ! "needs merge" msg &&
>
> Should we capture stderr from "git rm", too, to cover all bases?
Do you mean
	git rm blob >msg 2>&1

because we could later change our mind and send "needs merge" message to the standard error stream?

Jeff King· Jul 18, 2019, 21:43 UTC · re: Junio C Hamano · lore

Re: [PATCH] rm: resolving by removal is not a warning-worthy event

On Thu, Jul 18, 2019 at 02:07:23PM -0700, Junio C Hamano wrote:
Show 11 quoted lines
> >> +	git rm qfwfq >msg &&
> >> +	test_i18ngrep ! "needs merge" msg &&
> >
> > Should we capture stderr from "git rm", too, to cover all bases?
> 
> Do you mean
> 
> 	git rm blob >msg 2>&1
> 
> because we could later change our mind and send "needs merge"
> message to the standard error stream?
Yes, exactly.
-Peff

← back to recent threads