~bzr-pqm/bzr/bzr.dev

Bazaar Zen
==========

Grokking Bazaar
---------------

While Bazaar is similar to other VCS tools in many ways, there are
some important differences that are not necessarily obvious at first
glance.  This section attempts
to explain some of the things users need to know in order to "grok" Bazaar,
i.e. to deeply understand it.

Note: It isn't necessary to fully understand this section to use Bazaar.
You may wish to skim this section now and come back to it at a later time.

Understanding revision numbers
------------------------------

All revisions in the mainline of a branch have a simple increasing
integer. (First commit gets 1, 10th commit gets 10, etc.) This makes them
fairly natural to use when you want to say "grab the 10th revision from my
branch", or "fixed in revision 3050".

For revisions which have been merged into a branch, a dotted notation is used
(e.g., 3112.1.5). Dotted revision numbers have three numbers [#]_. The first
number indicates what mainline revision change is derived from. The second
number is the branch counter. There can be many branches derived from the
same revision, so they all get a unique number. The third number is the
number of revisions since the branch started. For example, 3112.1.5 is the
first branch from revision 3112, the fifth revision on that branch.

.. [#] Versions prior to bzr 1.2 used a slightly different algorithm.
   Some nested branches would get extra numbers (such as 1.1.1.1.1)
   rather than the simpler 3-number system.

Hierarchical history is good
----------------------------

Imagine a project with multiple developers contributing changes where
many changes consist of a series of commits. To give a concrete example,
consider the case where:

 * The tip of the project's trunk is revision 100.
 * Mary makes 3 changes to deliver feature X.
 * Bill makes 4 changes to deliver feature Y.

If the developers are working in parallel and using a traditional
centralized VCS approach, the project history will most likely be linear
with Mary's changes and Bill's changes interleaved. It might look like this::

  107: Add documentation for Y
  106: Fix bug found in testing Y
  105: Fix bug found in testing X
  104: Add code for Y
  103: Add documentation for X
  102: Add code and tests for X
  101: Add tests for Y
  100: ...

Many teams use this approach because their tools make branching and merging
difficult. As a consequence, developers update from and commit to the trunk
frequently, minimizing integration pain by spreading it over every commit.
If you wish, you can use Bazaar exactly like this. Bazaar does offer other
ways though that you ought to consider.

An alternative approach encouraged by distributed VCS tools is to create
feature branches and to integrate those when they are ready. In this case,
Mary's feature branch would look like this::

  103: Fix bug found in testing X
  102: Add documentation for X
  101: Add code and tests for X
  100: ...

And Bill's would look like this::

  104: Add documentation for Y
  103: Fix bug found in testing Y
  102: Add code for Y
  101: Add tests for Y
  100: ...

If the features were independent and you wanted to keep linear history,
the changes could be pushed back into the trunk in batches. (Technically,
there are several ways of doing that but that's beyond the scope of
this discussion.) The resulting history might look like this::

  107: Fix bug found in testing X
  106: Add documentation for X
  105: Add code and tests for X
  104: Add documentation for Y
  103: Fix bug found in testing Y
  102: Add code for Y
  101: Add tests for Y
  100: ...

While this takes a bit more effort to achieve, it has some advantages over
having revisions randomly intermixed. Better still though, branches can
be merged together forming a non-linear history. The result might look
like this::

  102: Merge feature X
       100.2.3: Fix bug found in testing X
       100.2.2: Add documentation for X
       100.2.1: Add code and tests for X
  101: Merge feature Y
       100.1.4: Add documentation for Y
       100.1.3: Fix bug found in testing Y
       100.1.2: Add code for Y
       100.1.1: Add tests for Y
  100: ...

Or more likely this::

  102: Merge feature X
       100.2.3: Fix bug
       100.2.2: Add documentation
       100.2.1: Add code and tests
  101: Merge feature Y
       100.1.4: Add documentation
       100.1.3: Fix bug found in testing
       100.1.2: Add code
       100.1.1: Add tests
  100: ...

This is considered good for many reasons:

 * It makes it easier to understand the history of a project.
   Related changes are clustered together and clearly partitioned.

 * You can easily collapse history to see just the commits on the mainline
   of a branch. When viewing the trunk history like this, you only see
   high level commits (instead of a large number of commits uninteresting
   at this level).

 * If required, it makes backing out a feature much easier.

 * Continuous integration tools can be used to ensure that
   all tests still pass before committing a merge to the mainline.
   (In many cases, it isn't appropriate to trigger CI tools after
   every single commit as some tests will fail during development.
   In fact, adding the tests first - TDD style - will guarantee it!)

In summary, the important points are:

  *Organize your work using branches.*

  *Integrate changes using merge.*

  *Ordered revision numbers and hierarchy make history easier to follow.*


Each branch has its own view of history
---------------------------------------

As explained above, Bazaar makes the distinction between:

 * mainline revisions, i.e. ones you committed in your branch, and

 * merged revisions, i.e. ones added as ancestors by committing a merge.

Each branch effectively has its own view of history, i.e. different
branches can give the same revision a different "local" revision number.
Mainline revisions always get allocated single number revision numbers
while merged revisions always get allocated dotted revision numbers.

To extend the example above, here's what the revision history of
Mary's branch would look like had she decided to merge the project
trunk into her branch after completing her changes::

  104: Merge mainline
       100.2.1: Merge feature Y
       100.1.4: Add documentation
       100.1.3: Fix bug found in testing
       100.1.2: Add code
       100.1.1: Add tests
  103: Fix bug found in testing X
  102: Add documentation for X
  101: Add code and tests for X
  100: ...

Once again, it's easy for Mary to look at just *her* top level of history
to see the steps she has taken to develop this change. In this context,
merging the trunk (and resolving any conflicts caused by doing that) is
just one step as far as the history of this branch is concerned.

It's important to remember that Bazaar is not changing history here, nor
is it changing the global revision identifiers. You can always use the
latter if you really want to. In fact, you can use the branch specific
revision numbers when communicating *as long as* you provide the branch
URL as context. (In many Bazaar projects, developers imply the central
trunk branch if they exchange a revision number without a branch URL.)

Merges do not change revision numbers in a branch, though they do
allocate local revision numbers to newly merged revisions. The only time
Bazaar will change revision numbers in a branch is when you explicitly
ask it to mirror another branch.

Note: Revisions are numbered in a stable way: if two branches have
the same revision in their mainline, all revisions in the ancestry of that
revision will have the same revision numbers. For example, if Alice and Bob's
branches agree on revision 10, they will agree on all revisions before
that.

Summary
-------

In general, if you follow the advice given earlier - organise
your work in branches and use merge to collaborate - you'll find Bazaar
generally does what you expect.

In coming chapters, we examine various ways of using Bazaar beginning with
the simplest: using Bazaar for personal projects.

..
   vim: ft=rst tw=74 ai