Using stacked branches
======================

Motivation
----------

If you are working on a project, and you have read access to whose
public repository but do not have write access to it, using stacked
branches to backup/publish your work onto the same host of the public
repository might be an option for you.

Other scenarios for stacked branch usage include experimental branches
and code hosting sites. For these scenarios, stacked branches are
ideal because of the benefits it provides.


What is a stacked branch?
-------------------------

A stacked branch is a branch that knows how to find revisions in
another branch (the stacked-on branch). Stacked branches store just
the unique revisions that are not in the stacked-on branch, making
them faster to create and more storage efficient. In these respects,
stacked branches are similar to shared repositories. However, stacked
branches have additional benefits:

* The new branch can be in a completely different location to the
  branch being stacked on.

* Deleting the stacked branch really deletes the revisions (rather
  than leaving them in a shared repository).

* Security is improved over shared repositories, because the stacked-on
  repository can be physically readonly to developers committing to stacked
  branches.


Creating a stacked branch
-------------------------

To create a stacked branch, use the ``stacked`` option of the branch command.
For example::

  bzr branch --stacked source-url my-dir

This will create ``my-dir`` as a stacked branch with no local revisions.
If it is defined, the public branch associated with ``source-url`` will be
used as the *stacked-on* location. Otherwise, ``source-url`` will be the
*stacked-on* location.


Creating a stacked checkout
---------------------------

Direct creation of a stacked checkout is expected to be supported soon.
In the meantime, a two step process is required:

1. Create a stacked branch as shown above.

2. Convert the branch into a checkout using either the ``reconfigure``
   or ``bind`` command.


Pushing a stacked branch
------------------------

Most changes on most projects build on an existing branch such as the
*development trunk* or *current stable* branch. Creating a new
branch stacked on one of these is easy to do using the ``push``
command like this::

  bzr push --stacked-on reference-url my-url

This creates a new branch at ``my-url`` that is stacked on ``reference-url``
and only contains the revisions in the current branch that are not already
in the branch at ``reference-url``. In particular, ``my-url`` and
``reference-url`` can be on the same host, and the ``--stacked-on`` option
can be used additionally to inform ``push`` to reference the
revisions in ``reference-url``. For example::

  bzr push --stacked-on bzr+ssh://host/project bzr+ssh://host/user/stacked-branch

This usage fits the scenario described in the Motivation section.

You can also use the ``--stacked`` option without specifying ``--stacked-on``.
This will automatically set the *stacked-on* location to the parent branch of
the branch you are pushing (or its ``public_location`` if configured).  For
example::

  bzr branch source-url my-dir
  cd my-dir
  (hack, hack, hack)
  bzr commit -m "fix bug"
  bzr push --stacked

You can combine ``bzr branch --stacked`` and ``bzr push --stacked`` to work on a
branch without downloading or uploading the whole history::

  bzr branch --stacked source-url my-dir
  cd my-dir
  (hack, hack, hack)
  bzr commit -m "fix bug"
  bzr push --stacked


Limitations of stacked branches
-------------------------------

The important thing to remember about a stacked branch is that the stacked-on
branch needs to be accessible for almost all operations.  This is not an issue
when both branches are local, or when both branches are on the same server and
the stacked-on location is a relative path.  But clearly a branch hosted on a
server with a stacked-on location of ``file:///...`` is not going to work for
anyone except the user that originally pushed it.  It's a good idea to configure
``public_location`` to help prevent that.

Similarly, because most of the history is stored in the stacked-on repository,
operations like ``bzr log`` can be slower when the stacked-on repository is
accessed via a network.

If a stacked branch is in a format older than 2a, you cannot commit to it due to
`bug 375013`_.

.. _bug 375013: https://bugs.launchpad.net/bzr/+bug/375013


Changing branch stacking
------------------------

Stacking of existing branches can be changed using the ``bzr reconfigure``
command to either stack on an existing branch, or to turn off stacking.
Be aware that when ``bzr reconfigure --unstacked`` is used, bzr will
copy all the referenced data from the stacked-on repository into the
previously stacked repository.  For large repositories this may take
considerable time and may substantially increase the size of the
repository.