Why is it bad to rewind a branch that has already been pushed out?

Why is it bad to rewind a branch that has already been pushed out?

[Junio C Hamano]

I was reading the tutorial and noticed that we say this:

    Also, don't use "git reset" on a publicly-visible branch that
    other developers pull from, as git will be confused by history
    that disappears in this way.

I do not think this is a good explanation.  For example, if we
do this:

(1) I build a series and push it out.

	---o---o---o---j

(2) Alice clones from me, and builds two commits on top of it.

	---o---o---o---j---a---a

(3) I rewind one and build a few, and push them out.

	---o---o---o...j
                    \
                     h---h---h---h

(4) Alice pulls from me again:

	---o---o---o---j---a---a---*
                    \             /
                     h---h---h---h

Contrary to the description, git will happily have Alice merge
between the two branches, and never gets confused.

Maybe I did not want to have 'j' because it was an incomplete
solution to some problem, and Alice may have fixed it up with
her changes, while I abandoned that approach I started with 'j',
and worked on something completely unrelated in the four 'h'
commits.  In such a case, the merge Alice would make would be
very sensible, and after she makes the merge if I pull from her,
the world will be perfect.  I started something with 'j' and
dropped the ball, Alice picked it up and perfected it while I
went on to work on something else with 'h'.  This would be a
perfect example of distributed parallel collaboration.  There is
nothing confused about it.

The case the rewinding becomes problematic is if the work done
in 'h' tries to solve the same problem as 'j' tried to solve in
a different way.  Then the merge forced on Alice would make her
pick between my previous attempt with her fixups (j+a) and my
second attempt (h).  If 'a' commits were to fix up what 'j'
started, presumably Alice already studied and knows enough about
the problem so she should be able to make an informed decision
to pick between what 'j+a' and 'h' do.

A lot worse case is if Alice's work is not at all related to
what 'j' wanted to do (she did not mean to pick up from where I
left off -- she just wanted to work on something different).
Then she would not be familiar enough with what 'j' and 'h'
tried to achieve, and I'd be forcing her to pick between the
two.  Of course if she can make the right decision, then again
that is a perfect example of distributed collaboration, but that
does not change the fact that I'd be forcing her to clean up my
mess.

So I am thinking about rewording the section like this.

Comments?

---

diff --git a/Documentation/tutorial.txt b/Documentation/tutorial.txt
index ea34189..5fc5be5 100644
--- a/Documentation/tutorial.txt
+++ b/Documentation/tutorial.txt
@@ -458,9 +458,9 @@ $ git reset --hard HEAD^ # reset your current branch and working
 Be careful with that last command: in addition to losing any changes
 in the working directory, it will also remove all later commits from
 this branch.  If this branch is the only branch containing those
-commits, they will be lost.  (Also, don't use "git reset" on a
-publicly-visible branch that other developers pull from, as git will
-be confused by history that disappears in this way.)
+commits, they will be lost.  Also, don't use "git reset" on a
+publicly-visible branch that other developers pull from, as it will
+force needless merges on other developers to clean up the history.
 
 The git grep command can search for strings in any version of your
 project, so

Re: Why is it bad to rewind a branch that has already been pushed out?

[Andy Parkins]

On Saturday 2007, February 03 06:40, Junio C Hamano wrote:

>     Also, don't use "git reset" on a publicly-visible branch that
>     other developers pull from, as git will be confused by history
>     that disappears in this way.
>
> I do not think this is a good explanation.  For example, if we
> do this:

I agree it's not a good explanation, I think it is good policy though.  
I love an oppourtunity to blather on and your post has given it to 
me :-)  I'm sure I'll be preaching to the converted though.

In the following I've assumed that both "h" and "a" commits are 
independent work, neither of which works on "j" (which isn't quite what 
you described).  It's not really appropriate for the tutorial, so I 
haven't really helped you.

-----------------
git's easy manipulation of branches encourages one to try to make 
history really tell the story of development.  It's more than just 
recording snapshots, there is information in the history itself - 
reasons for decisions can be recorded - separating and collecting 
commits into appropriate branches instead of one giant linear block.

As you say git is perfectly okay with the rewound branch and will do the 
Right Thing to preserve the changes.  However, it won't get the spirit 
of the changes right.

> (4) Alice pulls from me again:
>
> 	---o---o---o---j---a---a---*
>                   \             /
>                    h---h---h---h
>
> Contrary to the description, git will happily have Alice merge
> between the two branches, and never gets confused.

The problem is not that the working tree is wrong, it is that the 
history is wrong.  This sequence of development isn't best represented 
by a merge.  When we look at the history later, we'll see it as two 
branches, but that isn't true.

The graph shows that "j" was committed and then probably conflicted 
in "*" and fixed.  The problem is that you can't point to any non-merge 
commit that reverted/corrected "j" and explains why that wasn't a good 
idea - so the "story" that the history tells us is incomplete.  The 
correction was done just as a conflict resolution in "*".

Any of these would be correct:

1) "h" and "a" are parallel; "j" never happens

 ---o---o---o----a-----a----*
             \             /
              h---h---h---h

2) Alice corrects "j" with a "!j" revision - "!j" would then describe 
why "j" was wrong and what the fix was.  "h" would have to have nothing 
to do with the "j" topic in this case.

 ---o---o---o---j---!j---a---a---*
             \                  /
              h-----h-----h----h

Of course, both of these require information from the developer. (1) 
can't be done after the push has been done (because "j" is already "out 
there") and (2) requires that "h" know that Alice has fixed the problem 
and the Alice knows that the "j" problem needs fixing (and probably 
doesn't belong in Alice's branch anyway).

The only solution that keeps the history integrity, is the one that all 
the git manuals reccommend.

3) Don't rebase at all, correct the problem and record the correction.

  --o---o---o---j------a------a------*
                 \                  /
                  !j---h---h---h---h

That is - once a mistaken commit has been pushed out, it's there 
forever, accept that and don't rebase to before the last push.  
Sometimes it makes the history less pretty, but it makes it more right.  
Without "!j", no one would ever know what was wrong with "j".

In short - heed the warnings - you will get much better history if you 
don't rebase branches that have already been pushed upstream.
-----------------------

How about that as the explanation for the tutorial?  What do you mean I 
haven't simplified anything and have replaced one paragraph with a page 
worth of overly wordy ravings?  HOW DARE YOU!  :-)



Andy
-- 
Dr Andrew Parkins, M Eng (Hons), AMIEE
andyparkins@gmail.com

Re: Why is it bad to rewind a branch that has already been pushed out?

[Theodore Tso]

On Fri, Feb 02, 2007 at 10:40:49PM -0800, Junio C Hamano wrote:
> I do not think this is a good explanation.  For example, if we
> do this:
> 
> (1) I build a series and push it out.
> 
> 	---o---o---o---j
> 
> (2) Alice clones from me, and builds two commits on top of it.
> 
> 	---o---o---o---j---a---a
> 
> (3) I rewind one and build a few, and push them out.
> 
> 	---o---o---o...j
>                     \
>                      h---h---h---h
> 
> (4) Alice pulls from me again:
> 
> 	---o---o---o---j---a---a---*
>                     \             /
>                      h---h---h---h
> 
> Contrary to the description, git will happily have Alice merge
> between the two branches, and never gets confused.
> 
> Maybe I did not want to have 'j' because it was an incomplete
> solution to some problem....

Another scenario might be if 'j' was just a bad idea, and should have
never been committed in the first place.  Now if Alice builds upon it,
and at some later point Alice asks you to pull from her repository,
'j' will appear back in your repository.  So it's better to have a
changeset which simply undoes 'j' rathern than rewind, once you've
pushed out the repository.

The one exception to this is if 'j' contains something which might be
problematic from a copyright point of view, in which case you're
largely screwed if other people such as Alice has already pulled from
you and started basing changes on 'j'.

							- Ted

Re: Why is it bad to rewind a branch that has already been pushed out?

[Junio C Hamano]

Andy Parkins <andyparkins@gmail.com> writes:

> In the following I've assumed that both "h" and "a" commits are 
> independent work, neither of which works on "j" (which isn't quite what 
> you described).  It's not really appropriate for the tutorial, so I 
> haven't really helped you.
> ...
>> (4) Alice pulls from me again:
>>
>> 	---o---o---o---j---a---a---*
>>                   \             /
>>                    h---h---h---h
>>
>> Contrary to the description, git will happily have Alice merge
>> between the two branches, and never gets confused.
>
> The problem is not that the working tree is wrong, it is that the 
> history is wrong.  This sequence of development isn't best represented 
> by a merge.  When we look at the history later, we'll see it as two 
> branches, but that isn't true.
>
> The graph shows that "j" was committed and then probably conflicted 
> in "*" and fixed.  The problem is that you can't point to any non-merge 
> commit that reverted/corrected "j" and explains why that wasn't a good 
> idea - so the "story" that the history tells us is incomplete.  The 
> correction was done just as a conflict resolution in "*".

Actually, if you are assuming a, h, and j are unrelated, then
the merge done by Alice will _not_ revert 'j', so the history
will perfectly be fine.  The merge result will have a half-baked
work done with 'j', and everybody can build on top of.

Re: Why is it bad to rewind a branch that has already been pushed out?

[Robin Rosenberg]

Mention git-revert as an alternative to git-reset to revert changes.

Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com>
---

diff --git a/Documentation/tutorial.txt b/Documentation/tutorial.txt
index 5fc5be5..129c5c5 100644
--- a/Documentation/tutorial.txt
+++ b/Documentation/tutorial.txt
@@ -461,6 +461,8 @@ this branch.  If this branch is the only branch containing 
those
 commits, they will be lost.  Also, don't use "git reset" on a
 publicly-visible branch that other developers pull from, as it will
 force needless merges on other developers to clean up the history.
+If you need to undo changes that you have pushed, use gitlink:git-revert[1]
+instead.

 The git grep command can search for strings in any version of your
 project, so

Re: Why is it bad to rewind a branch that has already been pushed out?

[Andy Parkins]

On Saturday 2007, February 03 19:42, Junio C Hamano wrote:

> Actually, if you are assuming a, h, and j are unrelated, then
> the merge done by Alice will _not_ revert 'j', so the history
> will perfectly be fine.  The merge result will have a half-baked
> work done with 'j', and everybody can build on top of.

Absolutely true.

Your point is a strong one.  I think I'm still not thinking big enough 
with the distributed development concept.  Your point has made me 
consider that I've been overly concerned about not rebasing pushed 
commits.  Provided you're willing to correct conflicts in a merge; they 
aren't as bad as I initially thought - in fact they offer a further 
choice when creating project history.

Thanks Junio.



Andy
-- 
Dr Andrew Parkins, M Eng (Hons), AMIEE
andyparkins@gmail.com

Re: Why is it bad to rewind a branch that has already been pushed out?

[Junio C Hamano]

Robin Rosenberg <robin.rosenberg.lists@dewire.com> writes:

> Mention git-revert as an alternative to git-reset to revert changes.

Thanks; that makes sense.  Applied.

It is not a biggie but for future reference you might want to
check your KMail configuration to avoid the wrapped line below:

> diff --git a/Documentation/tutorial.txt b/Documentation/tutorial.txt
> index 5fc5be5..129c5c5 100644
> --- a/Documentation/tutorial.txt
> +++ b/Documentation/tutorial.txt
> @@ -461,6 +461,8 @@ this branch.  If this branch is the only branch containing 
> those
>  commits, they will be lost.  Also, don't use "git reset" on a

[PATCH] Documentation: add KMail in SubmittingPatches

[Michael]

Signed-off-by: Michael <barra_cuda@katamail.com>
---
 Documentation/SubmittingPatches |   19 ++++++++++++++++++-
 1 files changed, 18 insertions(+), 1 deletions(-)

diff --git a/Documentation/SubmittingPatches b/Documentation/SubmittingPatches
index ce85d06..a3dafa5 100644
--- a/Documentation/SubmittingPatches
+++ b/Documentation/SubmittingPatches
@@ -316,7 +316,6 @@ settings but I haven't tried, yet.
 	mail.identity.id?.compose_html		=> false
 
 
-
 Gnus
 ----
 
@@ -331,3 +330,21 @@ whitespaces (fatal in patches).  Running 'C-u g' to 
display the
 message in raw form before using '|' to run the pipe can work
 this problem around.
 
+
+KMail
+-----
+
+This should help you to submit patches inline using KMail.
+
+1) Prepare the patch as a text file.
+
+2) Click on New Mail.
+
+3) Go under "Options" in the Composer window and be sure that
+"Word wrap" is not set.
+
+4) Use Message -> Insert file... and insert the patch.
+
+5) Back in the compose window: add whatever other text you wish to the
+message, complete the addressing and subject fields, and press send.
+
-- 
1.4.4.2

Re: [PATCH] Documentation: add KMail in SubmittingPatches

[Junio C Hamano]

Michael <barra_cuda@katamail.com> writes:

> @@ -331,3 +330,21 @@ whitespaces (fatal in patches).  Running 'C-u g' to 
> display the
>  message in raw form before using '|' to run the pipe can work
>  this problem around.
>  
> +
> +KMail
> +-----
> +
> +This should help you to submit patches inline using KMail.
> +
> +1) Prepare the patch as a text file.
> +
> +2) Click on New Mail.
> +
> +3) Go under "Options" in the Composer window and be sure that
> +"Word wrap" is not set.
> +
> +4) Use Message -> Insert file... and insert the patch.
> +
> +5) Back in the compose window: add whatever other text you wish to the
> +message, complete the addressing and subject fields, and press send.
> +

This is quite interesting -- notice your own hunk header @@ ... @@ line?

I do not use KMail myself, so I cannot comment on the procedure,
but from the cursory look it *should* do the right thing.  Only
that it makes me a bit nervous to see your hunk header being
line wrapped.

Doubly interesting is that somebody appears to have added two
message IDs.

   Subject: [PATCH] Documentation: add KMail in SubmittingPatches
   Date:	Mon, 5 Feb 2007 14:27:32 +0100
   User-Agent: KMail/1.9.4
   Message-Id: <200702051427.32532.barra_cuda@katamail.com>
   Message-ID: <20070205132150.123659@dial-up-mi-449.lombardiacom.it>

Re: [PATCH] Documentation: add KMail in SubmittingPatches

[Michael]

Junio C Hamano:
> Michael <barra_cuda@katamail.com> writes:
> > @@ -331,3 +330,21 @@ whitespaces (fatal in patches).  Running 'C-u g' to
> > display the
> >  message in raw form before using '|' to run the pipe can work
> >  this problem around.
[...]
> This is quite interesting -- notice your own hunk header @@ ... @@ line?

I see. It's this account's fault: now I know because I've tried with
another account. I didn't know it had this glitch, sorry.

> I do not use KMail myself, so I cannot comment on the procedure,
> but from the cursory look it *should* do the right thing.  Only
> that it makes me a bit nervous to see your hunk header being
> line wrapped.
>
> Doubly interesting is that somebody appears to have added two
> message IDs.
>
>    Subject: [PATCH] Documentation: add KMail in SubmittingPatches
>    Date:	Mon, 5 Feb 2007 14:27:32 +0100
>    User-Agent: KMail/1.9.4
>    Message-Id: <200702051427.32532.barra_cuda@katamail.com>
>    Message-ID: <20070205132150.123659@dial-up-mi-449.lombardiacom.it>

This is my current account's fault too. So I used KMail correctly,
but my mail server did something to the message. I guess I should
use another account. Anyway, if some other KMail user has something
to say about this issue, it will make things clearer.

Bye.

Re: [PATCH] Documentation: add KMail in SubmittingPatches

[Jakub Narebski]

Michael wrote:

> This is my current account's fault too. So I used KMail correctly,
> but my mail server did something to the message. I guess I should
> use another account. Anyway, if some other KMail user has something
> to say about this issue, it will make things clearer.

I have send patches succesfully via KMail, using described procedure
(Options > Word wrap, Message > Insert File, move subject line to
Subject field, remove email headers unless with different From,...).
-- 
Jakub Narebski
Warsaw, Poland
ShadeHawk on #git

Re: [PATCH] Documentation: add KMail in SubmittingPatches

[Andy Parkins]

On Tuesday 2007 February 06 03:07, Junio C Hamano wrote:

> I do not use KMail myself, so I cannot comment on the procedure,
> but from the cursory look it *should* do the right thing.  Only
> that it makes me a bit nervous to see your hunk header being
> line wrapped.

I use Kmail for submitting patches, and it seems to work okay.  The important 
thing is to turn word wrap off as soon as you open it.  If you did this:

 * Save patch as draft (say with git-imap-send)
 * Open draft
 * Close draft
 * Open draft
 * Turn word wrap off
 * Send

You would have a badly wrapped patch.  The reason is that until you store the 
email again, kmail seems to know the difference between a hard LF that was in 
the original email and a LF that it inserted to wrap the lines.  So, as soon 
as you turn word wrap off all the soft LFs go away - perfect.  However, if 
you save without turning word wrap off, all those soft LFs become hard and 
the next time you open the email there is no easy way to get the untouched 
email back.

Michael's original instructions seem fine though.  I don't think his word wrap 
problem in this case is the work of KMail.



Andy

-- 
Dr Andy Parkins, M Eng (hons), MIEE
andyparkins@gmail.com