threads / discuss / 23942

rebase --continue confusion

Subject: rebase --continue confusion

## tl;dr

12 messages between May 30, 2010 and Jun 7, 2010.

replies: 11people: 8as markdown or json

Dale Rowley· May 30, 2010, 00:59 UTC · lore

I ran into a conflict while running 'git rebase branch1 branch2'. So I edited the file and resolved the conflicts, and then ran 'git add <file>; git rebase --continue'. This printed out a message 'No changes - did you forget to use git add?'. I thought 'No, I'm pretty sure I ran git add' and I assumed I had run into a bug in git because I didn't see how this conflict was different than any others I had successfully resolved. The next time this problem came up, the light finally came on and I realized that I should just run 'git rebase --skip' because when I resolved the conflicts, I had basically undone all changes that the patch would introduce.

OK, so there isn't a bug in git, but since then I've seen co-workers stumped by this same problem. So maybe it would help to clarify the message? Maybe something like "The index is in the same state as it was before the patch was applied - refusing to make an empty commit. Did you forget to use 'git add'? Or maybe you should use 'git rebase --skip'?"

Dale
Tim Visher· May 30, 2010, 12:44 UTC · re: Dale Rowley · lore

Re: rebase --continue confusion

On Sat, May 29, 2010 at 8:59 PM, Dale Rowley <ddrowley@gmail.com> wrote:
Show 15 quoted lines
> I ran into a conflict while running 'git rebase branch1 branch2'. So I edited
> the file and resolved the conflicts, and then ran 'git add <file>; git rebase
> --continue'. This printed out a message 'No changes - did you forget to use git
> add?'. I thought 'No, I'm pretty sure I ran git add' and I assumed I had run
> into a bug in git because I didn't see how this conflict was different than any
> others I had successfully resolved. The next time this problem came up, the
> light finally came on and I realized that I should just run 'git rebase --skip'
> because when I resolved the conflicts, I had basically undone all changes that
> the patch would introduce.
>
> OK, so there isn't a bug in git, but since then I've seen co-workers stumped by
> this same problem. So maybe it would help to clarify the message? Maybe
> something like "The index is in the same state as it was before the patch was
> applied - refusing to make an empty commit. Did you forget to use 'git add'? Or
> maybe you should use 'git rebase --skip'?"
2 things.
1. I agree that the message could be more informative, especially
given the context.  If the the patch looks exactly like the last patch
applied now that you've edited it during a rebase, it would _almost_
be safe to assume that you meant to skip and just didn't know you were
going to before hand.
2. `git status` should show you that you had nothing to commit, which
would completely explain why `git rebase --continue` threw out the
message it did.  I'm in the habit of running `git st` continually and
it often helps me catch this sort of thing where I thought I'd added
something to the index or forgot that I'd changed some file between
the last time I'd updated the index or something along those lines.
This is just sort of a recommendation on a way to modify your work
flow that may help you and your coworkers in the future.  YMMV.
-- 
In Christ,

Timmy V.

http://burningones.com/
http://five.sentenc.es/ - Spend less time on e-mail
skillzero@gmail.com· May 31, 2010, 00:14 UTC · re: Dale Rowley · lore

Re: rebase --continue confusion

On Sat, May 29, 2010 at 5:59 PM, Dale Rowley <ddrowley@gmail.com> wrote:
Show 15 quoted lines
> I ran into a conflict while running 'git rebase branch1 branch2'. So I edited
> the file and resolved the conflicts, and then ran 'git add <file>; git rebase
> --continue'. This printed out a message 'No changes - did you forget to use git
> add?'. I thought 'No, I'm pretty sure I ran git add' and I assumed I had run
> into a bug in git because I didn't see how this conflict was different than any
> others I had successfully resolved. The next time this problem came up, the
> light finally came on and I realized that I should just run 'git rebase --skip'
> because when I resolved the conflicts, I had basically undone all changes that
> the patch would introduce.
>
> OK, so there isn't a bug in git, but since then I've seen co-workers stumped by
> this same problem. So maybe it would help to clarify the message? Maybe
> something like "The index is in the same state as it was before the patch was
> applied - refusing to make an empty commit. Did you forget to use 'git add'? Or
> maybe you should use 'git rebase --skip'?"

I agree. I actually wasn't as smart as you when I saw that message in the past. I didn't know what the problem was and I didn't know about git rebase --skip (or I probably ignored it when I tried git rebase -h since skipping didn't sound like what I wanted at the time, but thanks for pointing it out if run into this in the future). I think I'd like git to just skip automatically in this case.

Eli Barzilay· Jun 6, 2010, 13:10 UTC · re: Dale Rowley · lore

Re: rebase --continue confusion

Dale Rowley <ddrowley@gmail.com> writes:
Show 7 quoted lines
>
> OK, so there isn't a bug in git, but since then I've seen co-workers
> stumped by this same problem. So maybe it would help to clarify the
> message? Maybe something like "The index is in the same state as it
> was before the patch was applied - refusing to make an empty
> commit. Did you forget to use 'git add'? Or maybe you should use
> 'git rebase --skip'?"

There are a number of these things that I asked about recently. Here's another similarly one:

  $ git add foo
  $ git status -s
  M  foo
  $ git commit --amend foo
  # On branch master
  # No changes
  $ git status -s
  M  foo
-- 
          ((lambda (x) (x x)) (lambda (x) (x x)))          Eli Barzilay:
                    http://barzilay.org/                   Maze is Life!
Jeff King· Jun 6, 2010, 22:18 UTC · re: Eli Barzilay · lore

Re: rebase --continue confusion

On Sun, Jun 06, 2010 at 09:10:44AM -0400, Eli Barzilay wrote:
Show 11 quoted lines
> There are a number of these things that I asked about recently.
> Here's another similarly one:
> 
>   $ git add foo
>   $ git status -s
>   M  foo
>   $ git commit --amend foo
>   # On branch master
>   # No changes
>   $ git status -s
>   M  foo

I'm confused. Is there some context for when you are issuing these commands? Because the "git commit --amend foo" should actually commit foo, and does for me.

-Peff
Eli Barzilay· Jun 6, 2010, 22:32 UTC · re: Jeff King · lore

Re: rebase --continue confusion

On Jun  6, Jeff King wrote:
Show 17 quoted lines
> On Sun, Jun 06, 2010 at 09:10:44AM -0400, Eli Barzilay wrote:
> 
> > There are a number of these things that I asked about recently.
> > Here's another similarly one:
> > 
> >   $ git add foo
> >   $ git status -s
> >   M  foo
> >   $ git commit --amend foo
> >   # On branch master
> >   # No changes
> >   $ git status -s
> >   M  foo
> 
> I'm confused. Is there some context for when you are issuing these
> commands?  Because the "git commit --amend foo" should actually
> commit foo, and does for me.

Heh, in that case it was more effective than I thought... My point in the previous posts was also about missing information (in that case, make `git add' tell you when adding it canceled previously added changes, and also make `git status' tell you if you're in the middle of a merge or rebase and in a clean state).

In any case, here's the prelude to the above:
  $ mkdir t; cd t; git init
  $ echo foo > foo; git add foo; git commit -m foo
  $ echo bar > foo; git commit -o foo -m bar
  $ echo foo > foo

(And as cooked as it seems, I posted it because I actually ran into it this morning.)

-- 
          ((lambda (x) (x x)) (lambda (x) (x x)))          Eli Barzilay:
                    http://barzilay.org/                   Maze is Life!
Jeff King· Jun 6, 2010, 22:46 UTC · re: Eli Barzilay · lore

Re: rebase --continue confusion

On Sun, Jun 06, 2010 at 06:32:58PM -0400, Eli Barzilay wrote:
Show 25 quoted lines
> > >   $ git add foo
> > >   $ git status -s
> > >   M  foo
> > >   $ git commit --amend foo
> > >   # On branch master
> > >   # No changes
> > >   $ git status -s
> > >   M  foo
> > 
> > I'm confused. Is there some context for when you are issuing these
> > commands?  Because the "git commit --amend foo" should actually
> > commit foo, and does for me.
> 
> Heh, in that case it was more effective than I thought...  My point in
> the previous posts was also about missing information (in that case,
> make `git add' tell you when adding it canceled previously added
> changes, and also make `git status' tell you if you're in the middle
> of a merge or rebase and in a clean state).
> 
> In any case, here's the prelude to the above:
> 
>   $ mkdir t; cd t; git init
>   $ echo foo > foo; git add foo; git commit -m foo
>   $ echo bar > foo; git commit -o foo -m bar
>   $ echo foo > foo

Ah, I see. Your problem has nothing to do with explicit pathnames (which I thought was the interesting bit from your snippet), but rather that you are amending it to the same as HEAD^.

Probably it would be helpful in the case of an amend to indicate what has happened (you have no changes, but it is not immediately obvious that you have no changes against HEAD^, not HEAD). We could even suggest "git reset HEAD^", which is probably what you want (the only other thing you could want is to create a commit with no changes, which we generally try to avoid).

-Peff
Eli Barzilay· Jun 6, 2010, 23:42 UTC · re: Jeff King · lore

Re: rebase --continue confusion

On Jun  6, Jeff King wrote:
Show 31 quoted lines
> On Sun, Jun 06, 2010 at 06:32:58PM -0400, Eli Barzilay wrote:
> 
> > > >   $ git add foo
> > > >   $ git status -s
> > > >   M  foo
> > > >   $ git commit --amend foo
> > > >   # On branch master
> > > >   # No changes
> > > >   $ git status -s
> > > >   M  foo
> > > 
> > > I'm confused. Is there some context for when you are issuing these
> > > commands?  Because the "git commit --amend foo" should actually
> > > commit foo, and does for me.
> > 
> > Heh, in that case it was more effective than I thought...  My point in
> > the previous posts was also about missing information (in that case,
> > make `git add' tell you when adding it canceled previously added
> > changes, and also make `git status' tell you if you're in the middle
> > of a merge or rebase and in a clean state).
> > 
> > In any case, here's the prelude to the above:
> > 
> >   $ mkdir t; cd t; git init
> >   $ echo foo > foo; git add foo; git commit -m foo
> >   $ echo bar > foo; git commit -o foo -m bar
> >   $ echo foo > foo
> 
> Ah, I see. Your problem has nothing to do with explicit pathnames (which
> I thought was the interesting bit from your snippet), but rather that
> you are amending it to the same as HEAD^.
Yes.
Show 6 quoted lines
> Probably it would be helpful in the case of an amend to indicate
> what has happened (you have no changes, but it is not immediately
> obvious that you have no changes against HEAD^, not HEAD). We could
> even suggest "git reset HEAD^", which is probably what you want (the
> only other thing you could want is to create a commit with no
> changes, which we generally try to avoid).

Yes, that sounds reasonable. (When I realized what happened I wondered why it didn't do the reset itself, but that would obviously be a bad idea.)

-- 
          ((lambda (x) (x x)) (lambda (x) (x x)))          Eli Barzilay:
                    http://barzilay.org/                   Maze is Life!
Jeff King· Jun 7, 2010, 00:41 UTC · re: Eli Barzilay · lore

Re: rebase --continue confusion

On Sun, Jun 06, 2010 at 07:42:40PM -0400, Eli Barzilay wrote:
Show 10 quoted lines
> > Probably it would be helpful in the case of an amend to indicate
> > what has happened (you have no changes, but it is not immediately
> > obvious that you have no changes against HEAD^, not HEAD). We could
> > even suggest "git reset HEAD^", which is probably what you want (the
> > only other thing you could want is to create a commit with no
> > changes, which we generally try to avoid).
> 
> Yes, that sounds reasonable.  (When I realized what happened I
> wondered why it didn't do the reset itself, but that would obviously
> be a bad idea.)

I'm not convinced this comes up often enough to really be worth worrying about, but the patch is quite trivial, so why not?

-- >8 --
Subject: [PATCH] commit: give advice on empty amend

We generally disallow empty commits with "git commit". The output produced by the wt_status functions is generally sufficient to explain what happened.

With --amend commits, however, things are a little more confusing. We would create an empty commit not if you actually have staged changes _now_, but if your staged changes match HEAD^. In this case, it is not immediately obvious why "git commit" claims no changes, but "git status" does not. Furthermore, we should point the user in the direction of git reset, which would eliminate the empty commit entirely.

Signed-off-by: Jeff King <peff@peff.net>
---
 builtin/commit.c |    7 +++++++
 1 files changed, 7 insertions(+), 0 deletions(-)
diff --git a/builtin/commit.c b/builtin/commit.c
index 4b2a468..10b09b9 100644
--- a/builtin/commit.c
+++ b/builtin/commit.c
@@ -48,6 +48,11 @@ static const char implicit_ident_advice[] =
 "\n"
 "    git commit --amend --author='Your Name <you@example.com>'\n";
 
+static const char empty_amend_advice[] =
+"You asked to amend the most recent commit, but doing so would make\n"
+"it empty. You can repeat your command with --allow-empty, or you can\n"
+"remove the commit entirely with \"git reset HEAD^\".\n";
+
 static unsigned char head_sha1[20];
 
 static char *use_message_buffer;
@@ -706,6 +711,8 @@ static int prepare_to_commit(const char *index_file, const char *prefix,
 	if (!commitable && !in_merge && !allow_empty &&
 	    !(amend && is_a_merge(head_sha1))) {
 		run_status(stdout, index_file, prefix, 0, s);
+		if (amend)
+			fputs(empty_amend_advice, stderr);
 		return 0;
 	}
 

-- 
1.7.1.458.g792cd.dirty
Junio C Hamano· May 31, 2010, 19:31 UTC · lore

Re: [PATCH] git-am: suggest what to do with superfluous patches

Jan Krüger <jk@jk.gs> writes:
Show 15 quoted lines
> Particularly in the context of rebase, conflicts frequently occur
> because the change in the patch to be applied was made obsolete by new
> upstream commits. In this case, solving the conflict effectively means
> skipping the patch. However, it's not always readily apparent that the
> patch needs to be skipped, and when people solve the conflict and try
> git rebase --continue, they get confronted with a message of
>
>   No changes - did you forget to use 'git add'?
>
> That's not very helpful if you did actually stage your changes and they
> happen to turn the patch into a no-op. This extends the message to point
> out what's going on.
>
> Signed-off-by: Jan Krüger <jk@jk.gs>
> ---

I think this is a change in a good direction; we _might_ want to allow this squelched with "advice.*" configuration, but my gut feeling is that it wouldn't probably matter much, as it is rather rare to trigger this.

Show 10 quoted lines
> diff --git a/git-am.sh b/git-am.sh
> index 87ffae2..43ea52c 100755
> --- a/git-am.sh
> +++ b/git-am.sh
> @@ -726,6 +726,8 @@ do
>  		resolved=
>  		git diff-index --quiet --cached HEAD -- && {
>  			echo "No changes - did you forget to use 'git add'?"
> +			echo "If there is nothing left to stage, chances are that something else"
> +			echo "already introduced the same changes; you might want to skip this patch."

The exact wording I'd let people to fight out, but I think this is probably better than Ramkumar's one that says "if you dropped". The user may not know that he is doing an equivalent of dropping as a side effect of the new base that had accepted the same change, and your message nudges the reader to realize that.

Ramkumar Ramachandra· May 31, 2010, 20:17 UTC · re: Junio C Hamano · lore

Re: [PATCH] git-am: suggest what to do with superfluous patches

Hi Junio,
Junio C Hamano <gitster@pobox.com> wrote:
Show 17 quoted lines
> Jan Krüger <jk@jk.gs> writes:
>> diff --git a/git-am.sh b/git-am.sh
>> index 87ffae2..43ea52c 100755
>> --- a/git-am.sh
>> +++ b/git-am.sh
>> @@ -726,6 +726,8 @@ do
>>               resolved=
>>               git diff-index --quiet --cached HEAD -- && {
>>                       echo "No changes - did you forget to use 'git add'?"
>> +                     echo "If there is nothing left to stage, chances are that something else"
>> +                     echo "already introduced the same changes; you might want to skip this patch."
>
> The exact wording I'd let people to fight out, but I think this is
> probably better than Ramkumar's one that says "if you dropped".  The user
> may not know that he is doing an equivalent of dropping as a side effect
> of the new base that had accepted the same change, and your message nudges
> the reader to realize that.

I didn't see this patch on the list- did Jan send it to the list? Yeah, the wording here is nicer, except you might also want to include an explicit note on using "$cmdline --skip".

-- Ram
Jan Krüger· May 31, 2010, 22:51 UTC · re: Ramkumar Ramachandra · lore

Re: [PATCH] git-am: suggest what to do with superfluous patches

Ramkumar Ramachandra <artagnon@gmail.com> wrote:
> I didn't see this patch on the list- did Jan send it to the list?
> Yeah, the wording here is nicer, except you might also want to include
> an explicit note on using "$cmdline --skip".

Yes, I did, but apparently something I changed in my mail setup a while ago caused the list server to silently drop mails sent to the list. I think I've fixed it now. Sorry for the confusion.

-Jan

← back to recent threads