how to fix the problem correctly?

how to fix the problem correctly?

[Eugene Sajine]

I have a following situation (in dev branch):

3 files:
1.txt, 2.txt, 3.txt

A -> B -> C -> D -> E
        |                     |
   Master               dev

Between B and C the version of file 1.txt was screwed up (let’s say it
was copied into git repo from other location where it was in state A
(CVS) and then committed by mistake in C, effectively reverting the
change of this file in commit B)
Commit C also had changes to files 2 and 3.
After this commits D and E did not have changes for file 1.txt, only 2
and 3 were touched.

When we tried to revert commit C – this lead to the code completely
messed up, with conflicts – so this seems to be not an option.
Rebase also doesn’t seem to help here

Only two variants I saw:
1. Do “git co A 1.txt” and commit the change as F
2. Do “git reset –soft master” and recommit all changes once again

Is there a better way?

PS interesting enough – CVS keywords helped us to notice the problem
as master state was imported from CVS.
In commit A file 1.txt had version ID 1.5 in commit B it was 1.6 ,
commit C was changing the line back to 1.5
Is there a way for git to help me to recognize this kind of issue if
there are no keywords?

DISCLAIMER: I don’t like keywords, and I don’t want them to be
implemented in git.


Thanks,
Eugene

Re: how to fix the problem correctly?

[Avery Pennarun]

On Wed, Apr 28, 2010 at 12:54 PM, Eugene Sajine <euguess@gmail.com> wrote:
> Between B and C the version of file 1.txt was screwed up (let’s say it
> was copied into git repo from other location where it was in state A
> (CVS) and then committed by mistake in C, effectively reverting the
> change of this file in commit B)
> Commit C also had changes to files 2 and 3.
> After this commits D and E did not have changes for file 1.txt, only 2
> and 3 were touched.
>
> When we tried to revert commit C – this lead to the code completely
> messed up, with conflicts – so this seems to be not an option.
> Rebase also doesn’t seem to help here
>
> Only two variants I saw:
> 1. Do “git co A 1.txt” and commit the change as F
> 2. Do “git reset –soft master” and recommit all changes once again
>
> Is there a better way?

Virtually any way that works is "correct."  It depends a bit on your goals.

Step 1 is certainly the easiest place to start.  If you're then
concerned about making sure your history never showed the mistake
(which is a lofty goal, though rarely very important), you could use
git rebase to 'squash' this new commit into C.  But rewriting history
in git has well-documented dangers, so you should be careful and read
the docs first.

> PS interesting enough – CVS keywords helped us to notice the problem
> as master state was imported from CVS.
> In commit A file 1.txt had version ID 1.5 in commit B it was 1.6 ,
> commit C was changing the line back to 1.5
> Is there a way for git to help me to recognize this kind of issue if
> there are no keywords?

Sadly, git doesn't have any magic features for detecting when someone
checks in something stupid :)   But 'git bisect' can be very helpful
in isolating which commit caused a particular problem.  Once you know
you have a problem, it's pretty easy to narrow it down that way.

Have fun,

Avery

Re: how to fix the problem correctly?

[Eugene Sajine]

>
> Virtually any way that works is "correct."  It depends a bit on your goals.
>
> Step 1 is certainly the easiest place to start.  If you're then
> concerned about making sure your history never showed the mistake
> (which is a lofty goal, though rarely very important), you could use
> git rebase to 'squash' this new commit into C.  But rewriting history
> in git has well-documented dangers, so you should be careful and read
> the docs first.

Good idea to clean up with rebase after step 1. I didn't think about
that, thanks! And yes I know about the restrictions;)

>
>> PS interesting enough – CVS keywords helped us to notice the problem
>> as master state was imported from CVS.
>> In commit A file 1.txt had version ID 1.5 in commit B it was 1.6 ,
>> commit C was changing the line back to 1.5
>> Is there a way for git to help me to recognize this kind of issue if
>> there are no keywords?
>
> Sadly, git doesn't have any magic features for detecting when someone
> checks in something stupid :)   But 'git bisect' can be very helpful
> in isolating which commit caused a particular problem.  Once you know
> you have a problem, it's pretty easy to narrow it down that way.

yes, it is very hard to be fool-proof. I just got an idea:
In this particular case some script scanning for pairs of commits
where blob SHA-1 changes
are like below (for the same path) and warns about such occurrences.

commit 1111
... blobaaaaaa... blobbbbbbb... path

commit 3333

... blobbbbbbb... blobaaaaaaa... path

Unless the commit 3333 is a real revert, this might be suspicious, isn't it?

I don't know if it has any real sense or use, but i will try to create
such thing at least as an exercise...

Thanks,
Eugene