What is the purpose of `rmbranch` command?

[Alexander Belchenko]

I thought `rmbranch` should delete a branch from remote server.

It does delete branch (e.g. from shared repository) but leaves .bzr
control directory there intact.

According to `rmbranch` help it should keep any working tree or
repository in place. OK, that's good.

But if there is neither working tree nor repository, why don't delete an
empty control directory?

[Jelmer Vernooij-2]

Hi Alexander,

On Tue, May 15, 2012 at 06:20:12PM +0300, Alexander Belchenko wrote:
> I thought `rmbranch` should delete a branch from remote server.

> It does delete branch (e.g. from shared repository) but leaves .bzr
> control directory there intact.
Right, it just deletes the branch, not the control directory.

> According to `rmbranch` help it should keep any working tree or
> repository in place. OK, that's good.

> But if there is neither working tree nor repository, why don't
> delete an empty control directory?
A control directory can have multiple branches. rmbranch is useful
mostly for colocated branches.

I don't think having a separate command for removing a control
directory would be particularly useful - it's already possible to just
remove the directory using e.g. the shell.

Cheers,

Jelmer

[Vincent Ladeuil-2]

>>>>> Jelmer Vernooij <jelmer@...> writes:

    > Hi Alexander,
    > On Tue, May 15, 2012 at 06:20:12PM +0300, Alexander Belchenko wrote:
    >> I thought `rmbranch` should delete a branch from remote server.

    >> It does delete branch (e.g. from shared repository) but leaves .bzr
    >> control directory there intact.
    > Right, it just deletes the branch, not the control directory.

    >> According to `rmbranch` help it should keep any working tree or
    >> repository in place. OK, that's good.

    >> But if there is neither working tree nor repository, why don't
    >> delete an empty control directory?
    > A control directory can have multiple branches. rmbranch is useful
    > mostly for colocated branches.

    > I don't think having a separate command for removing a control
    > directory would be particularly useful - it's already possible to just
    > remove the directory using e.g. the shell.

I don't think Alexander asked for a separate command. Instead, AIUI, his
point is that it makes little sense to remove the branch and leave an
otherwise *empty* control directory (not to mention other possible bugs
when trying to interact with this empty directory).

@bialix: worth filing a bug so we don't forget this.

         Vincent

[Jelmer Vernooij-2]

On Wed, May 16, 2012 at 09:01:52AM +0200, Vincent Ladeuil wrote:

> >>>>> Jelmer Vernooij <jelmer@...> writes:
>
>     > Hi Alexander,
>     > On Tue, May 15, 2012 at 06:20:12PM +0300, Alexander Belchenko wrote:
>     >> I thought `rmbranch` should delete a branch from remote server.
>
>     >> It does delete branch (e.g. from shared repository) but leaves .bzr
>     >> control directory there intact.
>     > Right, it just deletes the branch, not the control directory.
>
>     >> According to `rmbranch` help it should keep any working tree or
>     >> repository in place. OK, that's good.
>
>     >> But if there is neither working tree nor repository, why don't
>     >> delete an empty control directory?
>     > A control directory can have multiple branches. rmbranch is useful
>     > mostly for colocated branches.
>
>     > I don't think having a separate command for removing a control
>     > directory would be particularly useful - it's already possible to just
>     > remove the directory using e.g. the shell.
>
> I don't think Alexander asked for a separate command. Instead, AIUI, his
> point is that it makes little sense to remove the branch and leave an
> otherwise *empty* control directory (not to mention other possible bugs
> when trying to interact with this empty directory).

This isn't always the case though - there can be other colocated
branches, a repository or a working tree. We have no way to remove a
repository for example.

As far as I understand this would unversion the current directory, which
would seem a bit odd. Am I misunderstanding something?

Cheers,

Jelmer

[John Arbash Meinel]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

...

>> I don't think Alexander asked for a separate command. Instead,
>> AIUI, his point is that it makes little sense to remove the
>> branch and leave an otherwise *empty* control directory (not to
>> mention other possible bugs when trying to interact with this
>> empty directory).
> This isn't always the case though - there can be other colocated
> branches, a repository or a working tree. We have no way to remove
> a repository for example.
>
> As far as I understand this would unversion the current directory,
> which would seem a bit odd. Am I misunderstanding something?
>
> Cheers,
>
> Jelmer
>

bzr branch --no-tree lp:bzr target
bzr rmbranch target

Arguably, there should be no 'target/' directory left. Certainly that
is what people expect.

John
=:->
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Cygwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAk+zmfMACgkQJdeBCYSNAAOTjwCdGiLyQM9W9RAzNM7oRiFgBn67
UgYAoJUO/ewAsxyGuGpGj/VALDF0PrwN
=b4F6
-----END PGP SIGNATURE-----

[Vincent Ladeuil-2]

>>>>> Jelmer Vernooij <jelmer@...> writes:

<snip/>

    >> I don't think Alexander asked for a separate command. Instead, AIUI, his
    >> point is that it makes little sense to remove the branch and leave an
    >> otherwise *empty* control directory (not to mention other possible bugs
    >> when trying to interact with this empty directory).

    > This isn't always the case though - there can be other colocated
    > branches, a repository or a working tree. We have no way to remove a
    > repository for example.

But in those cases, the control directory (we're all talking about
'.bzr' not '.bzr/branch' right ?) is therefore not empty.

I don't think bialix is arguing about those valid cases.

  Vincent

[Jelmer Vernooij-12]

On Wed, May 16, 2012 at 02:13:39PM +0200, John Arbash Meinel wrote:

> >> I don't think Alexander asked for a separate command. Instead,
> >> AIUI, his point is that it makes little sense to remove the
> >> branch and leave an otherwise *empty* control directory (not to
> >> mention other possible bugs when trying to interact with this
> >> empty directory).
> > This isn't always the case though - there can be other colocated
> > branches, a repository or a working tree. We have no way to remove
> > a repository for example.
> >
> > As far as I understand this would unversion the current directory,
> > which would seem a bit odd. Am I misunderstanding something?
> bzr branch --no-tree lp:bzr target
> bzr rmbranch target
>
> Arguably, there should be no 'target/' directory left. Certainly that
> is what people expect.

What Vincent and Alexander are arguing is that in that case the
target/ directory should be kept around - it still contains a
repository.

I don't like the idea of automatically removing the control directory
in the case when only .bzr/branch exists (and no .bzr/checkout or
.bzr/repository) since it makes the behaviour of "bzr rmbranch"
unpredictable.

Cheers,

Jelmer

[Jelmer Vernooij-12]

On Wed, May 16, 2012 at 02:17:24PM +0200, Vincent Ladeuil wrote:

> >>>>> Jelmer Vernooij <jelmer@...> writes:
>
> <snip/>
>
>     >> I don't think Alexander asked for a separate command. Instead, AIUI, his
>     >> point is that it makes little sense to remove the branch and leave an
>     >> otherwise *empty* control directory (not to mention other possible bugs
>     >> when trying to interact with this empty directory).
>
>     > This isn't always the case though - there can be other colocated
>     > branches, a repository or a working tree. We have no way to remove a
>     > repository for example.
>
> But in those cases, the control directory (we're all talking about
> '.bzr' not '.bzr/branch' right ?) is therefore not empty.

That makes the behaviour of "bzr rmbranch" fairly unpredictable
though.

I can see the argument for removing a control directory if it's
completely empty, but I think we should be more consistent about that
- "bzr remove-tree" should do the same, and we should have a way to
remove a repository from a control directory.

Cheers,

Jelmer

[Vincent Ladeuil-2]

>>>>> Jelmer Vernooij <jelmer@...> writes:

    > On Wed, May 16, 2012 at 02:17:24PM +0200, Vincent Ladeuil wrote:
    >> >>>>> Jelmer Vernooij <jelmer@...> writes:
    >>
    >> <snip/>
    >>
    >> >> I don't think Alexander asked for a separate command. Instead, AIUI, his
    >> >> point is that it makes little sense to remove the branch and leave an
    >> >> otherwise *empty* control directory (not to mention other possible bugs
    >> >> when trying to interact with this empty directory).
    >>
    >> > This isn't always the case though - there can be other colocated
    >> > branches, a repository or a working tree. We have no way to remove a
    >> > repository for example.
    >>
    >> But in those cases, the control directory (we're all talking about
    >> '.bzr' not '.bzr/branch' right ?) is therefore not empty.
    > That makes the behaviour of "bzr rmbranch" fairly unpredictable
    > though.

Hmm, I don't fully agree but better understand your point.

    > I can see the argument for removing a control directory if it's
    > completely empty, but I think we should be more consistent about that
    > - "bzr remove-tree" should do the same, and we should have a way to
    > remove a repository from a control directory.

I agree with that too (with added controls in the repository case that
no branches uses the repository).

  Vincent

[Alexander Belchenko]

John Arbash Meinel пишет:

> ....
>
>>> I don't think Alexander asked for a separate command. Instead,
>>> AIUI, his point is that it makes little sense to remove the
>>> branch and leave an otherwise *empty* control directory (not to
>>> mention other possible bugs when trying to interact with this
>>> empty directory).
>> This isn't always the case though - there can be other colocated
>> branches, a repository or a working tree. We have no way to remove
>> a repository for example.
>>
>> As far as I understand this would unversion the current directory,
>> which would seem a bit odd. Am I misunderstanding something?
>>
>> Cheers,
>>
>> Jelmer
>>
>
> bzr branch --no-tree lp:bzr target
> bzr rmbranch target
>
> Arguably, there should be no 'target/' directory left. Certainly that
> is what people expect.

Yes, that was my intuitive expectation.

I'm OK with any behavior and don't claim current behavior as bug, but
I'd like to understand the intent of `rmbranch` command.

I see why Jelmer said that it was only useful for colocated branches.

My question arose because I tried to remove branch from bzr server
(bzr:// protocol + Trac) with that command, and I found that `rmbranch`
left almost everything on the server. That was confusing. (of course I
was able to inspect what's left on disk to understood what'd happened).

Now I better understand why other people confused, see
http://stackoverflow.com/questions/10474291/remove-a-parent-branch-in-bazaar/

I can only imagine a simple way to improve the current status-quo:
print something to the user when `bzr rmbranch URL:/to/branch` command
finishes, something about

`URL:/to/branch` is no more valid bzr branch, but still present on the disk.

In general, beyond colocated usage, this command is almost useless.

And I should note that for bzr-colo users it's useless at all, and
better to use special colo-prune (colo-delete) or qprune (qdelete) GUI
in all cases.

Thanks.

[Stefan Monnier]

> I don't think having a separate command for removing a control
> directory would be particularly useful - it's already possible to just
> remove the directory using e.g. the shell.

There are plenty of remote Bzr servers that do not offer access via
a shell.  I really wish I could rename and delete branches (and the
corresponding directories) without needing access to a shell.


        Stefan

[Aaron Bentley-4]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 12-05-16 11:44 AM, Stefan Monnier wrote:
>> I don't think having a separate command for removing a control
>> directory would be particularly useful - it's already possible
>> to just remove the directory using e.g. the shell.
>
> There are plenty of remote Bzr servers that do not offer access via
> a shell.  I really wish I could rename and delete branches (and the
> corresponding directories) without needing access to a shell.

There's always hitchhiker, and for some sites, an sftp client works too.

Aaron
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAk+z38QACgkQ0F+nu1YWqI3Q9gCfXrmAYn3o8hnyIQlMRQW9zhGZ
ekIAn1m/Z2MRUBL3Wsmfp8YF72GNUVuT
=neEm
-----END PGP SIGNATURE-----

[Stefan Monnier]

>>> I don't think having a separate command for removing a control
>>> directory would be particularly useful - it's already possible
>>> to just remove the directory using e.g. the shell.
>> There are plenty of remote Bzr servers that do not offer access via
>> a shell.  I really wish I could rename and delete branches (and the
>> corresponding directories) without needing access to a shell.
> There's always hitchhiker

Ha!  Just what I needed, thanks!


        Stefan