rebase --onto might "loose" commits

rebase --onto might "loose" commits

[Björn Steinbrink]

[-- Attachment #1: Type: text/plain, Size: 428 bytes --]

Hi,

when there's nothing to rebase (ie. upstream..branch is empty), rebase
fails to find any commits to rebase and correctly says "Nothing to do",
but when --onto is given, we already did a "reset --hard" to newbase, so
it already _did_ something.

As rebase also shows a "fatal: Not a range", during the operation, I
assume that this is an actual bug and not just a plain user error.

A test script is attached.

Thank,
Björn

[-- Attachment #2: test.sh --]
[-- Type: application/x-sh, Size: 616 bytes --]

Re: rebase --onto might "loose" commits

[Johannes Schindelin]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 1034 bytes --]

Hi,

On Tue, 19 Feb 2008, Björn Steinbrink wrote:

> when there's nothing to rebase (ie. upstream..branch is empty), rebase 
> fails to find any commits to rebase and correctly says "Nothing to do", 
> but when --onto is given, we already did a "reset --hard" to newbase, so 
> it already _did_ something.

Yes, it did something.  But if you had that:

A - B - C - D - E
  \
    F

your HEAD was E, and you said "git rebase --onto F E" what exactly do you 
want it to do?  There is no commit between E and E, so it rebases 
_nothing_ onto F.  Which means that F should be your new state.

Unless I am missing something critical in your mail.

> As rebase also shows a "fatal: Not a range", during the operation, I 
> assume that this is an actual bug and not just a plain user error.

That message is probably a bug, then.

Ciao,
Dscho

P.S.: I was being corrected some time ago on the same typo: "to lose" 
means to get rid of something unintentionally, "to loose" does not exist, 
and "to loosen" means to make something less tight.

Re: rebase --onto might "loose" commits

[Björn Steinbrink]

Hi,

On 2008.02.19 11:05:40 +0000, Johannes Schindelin wrote:
> On Tue, 19 Feb 2008, Björn Steinbrink wrote:
> 
> > when there's nothing to rebase (ie. upstream..branch is empty), rebase 
> > fails to find any commits to rebase and correctly says "Nothing to do", 
> > but when --onto is given, we already did a "reset --hard" to newbase, so 
> > it already _did_ something.
> 
> Yes, it did something.  But if you had that:
> 
> A - B - C - D - E
>   \
>     F
> 
> your HEAD was E, and you said "git rebase --onto F E" what exactly do you
> want it to do?  There is no commit between E and E, so it rebases
> _nothing_ onto F.  Which means that F should be your new state.

Strictly, it's correct, but the "Nothing to do" message is a bit
misleading (IMHO) and the error message made me think, that it actually
didn't want to do anything.

I'm a bit unsure about rebase being degraded to a "reset --hard" in this
case is a good idea. Might be a nice user-protection to make rebase
abort when there's nothing to rebase and --onto is given. But I don't
care that much.

Thanks,
Björn

> P.S.: I was being corrected some time ago on the same typo: "to lose" 
> means to get rid of something unintentionally, "to loose" does not exist, 
> and "to loosen" means to make something less tight.

Ah yeah, one of my preferred typos. There's about a 50% chance for me to
get that right. Although it already felt wrong when I read over the
mail, for some unknown reason I couldn't put my finger on the typo. ;-)

Re: rebase --onto might "loose" commits

[Nicolas Pitre]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 1515 bytes --]

On Tue, 19 Feb 2008, Björn Steinbrink wrote:

> Hi,
> 
> On 2008.02.19 11:05:40 +0000, Johannes Schindelin wrote:
> > On Tue, 19 Feb 2008, Björn Steinbrink wrote:
> > 
> > > when there's nothing to rebase (ie. upstream..branch is empty), rebase 
> > > fails to find any commits to rebase and correctly says "Nothing to do", 
> > > but when --onto is given, we already did a "reset --hard" to newbase, so 
> > > it already _did_ something.
> > 
> > Yes, it did something.  But if you had that:
> > 
> > A - B - C - D - E
> >   \
> >     F
> > 
> > your HEAD was E, and you said "git rebase --onto F E" what exactly do you
> > want it to do?  There is no commit between E and E, so it rebases
> > _nothing_ onto F.  Which means that F should be your new state.
> 
> Strictly, it's correct, but the "Nothing to do" message is a bit
> misleading (IMHO) and the error message made me think, that it actually
> didn't want to do anything.

Yes, the message should probably be changed in that case.

> I'm a bit unsure about rebase being degraded to a "reset --hard" in this
> case is a good idea. Might be a nice user-protection to make rebase
> abort when there's nothing to rebase and --onto is given. But I don't
> care that much.

The "reset --hard" in that case is really the correct behavior.  Rebase 
might be used like 'git pull --rebase', and if your local branch 
contains no commits of your own then the 'git reset --hard' is really 
what needs to happen.

The "user protection" lives in the reflog.


Nicolas