Problems switching branches

Problems switching branches

[Maaartin]

I sometimes run in a problem similar to
http://kerneltrap.org/mailarchive/git/2008/10/15/3667644/thread
There are some ignored files which I want neither track nor throw away; I'm just 
happy to have them and keep them out of version control.

Unfortunately, there weren't ignored in the old branch. I'd be quite happy with 
non-destructive switching like "checkout everything what doesn't overwrite an 
untracked file", so I would end in the old branch with a dirty working tree. Is 
it possible?

Re: Problems switching branches

[Maaartin]

Maaartin <grajcar1 <at> seznam.cz> writes:

> 
> I sometimes run in a problem similar to
> http://kerneltrap.org/mailarchive/git/2008/10/15/3667644/thread
> There are some ignored files which I want neither track nor throw away; I'm 
just 
> happy to have them and keep them out of version control.
> 
> Unfortunately, there weren't ignored in the old branch. I'd be quite happy 
with 
> non-destructive switching like "checkout everything what doesn't overwrite an 
> untracked file", so I would end in the old branch with a dirty working tree. 
Is 
> it possible?

No answer? In the meantime I've found a simple workaround: Checkout the old 
branch in another directory, modify .gitignore, commit and retry. But this is 
far from cool. Am I doing something stupid and nobody else ran into this?

Re: Problems switching branches

[Nguyen Thai Ngoc Duy]

On Thu, Nov 4, 2010 at 7:23 AM, Maaartin <grajcar1@seznam.cz> wrote:
> Maaartin <grajcar1 <at> seznam.cz> writes:
>
>>
>> I sometimes run in a problem similar to
>> http://kerneltrap.org/mailarchive/git/2008/10/15/3667644/thread
>> There are some ignored files which I want neither track nor throw away; I'm
> just
>> happy to have them and keep them out of version control.
>>
>> Unfortunately, there weren't ignored in the old branch. I'd be quite happy
> with
>> non-destructive switching like "checkout everything what doesn't overwrite an
>> untracked file", so I would end in the old branch with a dirty working tree.
> Is
>> it possible?
>
> No answer?

Which means nobody is interested in. Well, not really.

I also get irritated by a similar situation, where the untracked files
have the same content as the to-be-checked-out files. I have been
tempted (but never got around) to make git compare the in-index
content and the untracked file, if it's the same, no need to abort the
checkout process.

But your approach may be better. Yes, I think it's possible. Any
suggestion for checkout's new argument? --no-overwrite-untracked seems
too long.
-- 
Duy

Re: Problems switching branches

[Maaartin-1]

On 10-11-04 15:39, Nguyen Thai Ngoc Duy wrote:
> On Thu, Nov 4, 2010 at 7:23 AM, Maaartin <grajcar1@seznam.cz> wrote:
>> Maaartin <grajcar1 <at> seznam.cz> writes:
>>
>>>
>>> I sometimes run in a problem similar to
>>> http://kerneltrap.org/mailarchive/git/2008/10/15/3667644/thread
>>> There are some ignored files which I want neither track nor throw away; I'm
>> just
>>> happy to have them and keep them out of version control.
>>>
>>> Unfortunately, there weren't ignored in the old branch. I'd be quite happy
>> with
>>> non-destructive switching like "checkout everything what doesn't overwrite an
>>> untracked file", so I would end in the old branch with a dirty working tree.
>> Is
>>> it possible?
>>
>> No answer?
> 
> Which means nobody is interested in. Well, not really.
> 
> I also get irritated by a similar situation, where the untracked files
> have the same content as the to-be-checked-out files. I have been
> tempted (but never got around) to make git compare the in-index
> content and the untracked file, if it's the same, no need to abort the
> checkout process.

I was asked to provide a working example of the happening, but I haven't
managed to reproduce it yet. However, it's not very rare (it just never
happen when I need it).

> But your approach may be better. Yes, I think it's possible. Any
> suggestion for checkout's new argument? --no-overwrite-untracked seems
> too long.

I would go even further: a switch called "ignorant" or "lenient"
allowing to always switch branches in a non-destructible way. All files
normally causing abort would be left unmodified, so you could do
git checkout --ignorant forth; git checkout back
and would (assuming you started in branch "back") land in the original
state without loosing anything. Of course, this means, that the ignorant
checkout doesn't lead you into a clean state, but that's why I'd like to
use a switch instead of making it the default. :)

I may be talking non-sense as I'm quite inexperienced user, however I'd
love an easier way for switching branches. Quite often, I'd like to put
a modified file onto a different branch. This sounds probably strange,
but my work is really quite chaotic in this respect at the moment and
I'd like to organize it better by using a couple of (quite similar)
branches (with a lifespan of a few days at most).

Re: Problems switching branches

[Nguyen Thai Ngoc Duy]

On Fri, Nov 5, 2010 at 1:36 PM, Maaartin-1 <grajcar1@seznam.cz> wrote:
>> But your approach may be better. Yes, I think it's possible. Any
>> suggestion for checkout's new argument? --no-overwrite-untracked seems
>> too long.
>
> I would go even further: a switch called "ignorant" or "lenient"
> allowing to always switch branches in a non-destructible way. All files
> normally causing abort would be left unmodified, so you could do
> git checkout --ignorant forth; git checkout back
> and would (assuming you started in branch "back") land in the original
> state without loosing anything. Of course, this means, that the ignorant
> checkout doesn't lead you into a clean state, but that's why I'd like to
> use a switch instead of making it the default. :)

One thing to consider. If there are conflicts, I don't think we should
allow this "ignorant" mode. That would mess up work tree in a bad way.
And because it would leave worktree in a dirty state, maybe
--dirty-worktree is suggestive.
-- 
Duy

Re: Problems switching branches

[Neal Kreitzinger]

"Maaartin-1" <grajcar1@seznam.cz> wrote in message 
news:4CD3A5D9.6070802@seznam.cz...
> On 10-11-04 15:39, Nguyen Thai Ngoc Duy wrote:
>> On Thu, Nov 4, 2010 at 7:23 AM, Maaartin <grajcar1@seznam.cz> wrote:
>>> Maaartin <grajcar1 <at> seznam.cz> writes:
>>>
>>>>
>>>> I sometimes run in a problem similar to
>>>> http://kerneltrap.org/mailarchive/git/2008/10/15/3667644/thread
>>>> There are some ignored files which I want neither track nor throw away; 
>>>> I'm
>>> just
>>>> happy to have them and keep them out of version control.
>>>>
>>>> Unfortunately, there weren't ignored in the old branch. I'd be quite 
>>>> happy
>>> with
>>>> non-destructive switching like "checkout everything what doesn't 
>>>> overwrite an
>>>> untracked file", so I would end in the old branch with a dirty working 
>>>> tree.
>>> Is
>>>> it possible?
>>>
>>> No answer?
>>
>> Which means nobody is interested in. Well, not really.
>>
>> I also get irritated by a similar situation, where the untracked files
>> have the same content as the to-be-checked-out files. I have been
>> tempted (but never got around) to make git compare the in-index
>> content and the untracked file, if it's the same, no need to abort the
>> checkout process.
>
> I was asked to provide a working example of the happening, but I haven't
> managed to reproduce it yet. However, it's not very rare (it just never
> happen when I need it).
>
>> But your approach may be better. Yes, I think it's possible. Any
>> suggestion for checkout's new argument? --no-overwrite-untracked seems
>> too long.
>
> I would go even further: a switch called "ignorant" or "lenient"
> allowing to always switch branches in a non-destructible way. All files
> normally causing abort would be left unmodified, so you could do
> git checkout --ignorant forth; git checkout back
> and would (assuming you started in branch "back") land in the original
> state without loosing anything. Of course, this means, that the ignorant
> checkout doesn't lead you into a clean state, but that's why I'd like to
> use a switch instead of making it the default. :)
>
> I may be talking non-sense as I'm quite inexperienced user, however I'd
> love an easier way for switching branches. Quite often, I'd like to put
> a modified file onto a different branch. This sounds probably strange,
> but my work is really quite chaotic in this respect at the moment and
> I'd like to organize it better by using a couple of (quite similar)
> branches (with a lifespan of a few days at most).
>

You could use git-stash in combination with git-clean.  Let's say BranchA 
has FileA (untracked), and BranchB has FileA (tracked):
git branch = BranchA
git stash = this will store your dirty working tree in a commit 
behind-the-scenes, and store your index in a commit behind-the-scenes, and 
then run a git-reset --hard behind the scenes.
git status = is FileA still 'untracked'?
    if so, git clean -f to remove it.  don't worry, its in your stash you 
just made.
git clean -f
git checkout BranchB
now you have FileA from BranchB
git checkout BranchA
git stash apply, (use git stash apply --index if you want the index back 
also)
now you have BranchA and its untracked FileA

Using the stash and clean will also keep various untracked files from 
different lines-of-development from accumulating in your worktree.  "Keep 
track" of your untracked files by stashing them away and cleaning them up, 
so to speak.  This is actually pretty straightforwarad and easy once you get 
used to it, IMHO.

v/r,
Neal