Lukáš Lalinský

Many people dislike the directory-per-branch concept that Bazaar uses. What they don't realize though, is that this doesn't mean you need to have a working tree for each branch. You can very easily simulate cheap Git-style branches, but with some added flexibility. Checkouts are a fairly well known feature of Bazaar, but people mostly associate it with the centralized workflow (i.e. checking out remote branches). This is not the only use case for them.

When I work on larger projects, where I need multiple branches, I usually have a directory structure like this:

• "project"
  • "branches"
    • "branchA"
    • "branchB"
    • "trunk"
  • "work"

In this example, "project" is a shared repository. It contains revisions for project's branches on a single place. The repository is created with the --no-trees option, so that working trees are not automatically for new branches. All the branches I need to work with are located in "project/branches/XXX". Thanks to the DAG model, they represent nothing more than pointers to the "head" revision in the repository, so they are pretty cheap to create.

My development happens in "project/work", which is a lightweight checkout to one of the branches. This means that it doesn't contain anything else but information about the state of the working tree and a pointer to the branch. For any operation, Bazaar will use the branch it points to instead.

I'll use QBzr as an example how to set this up:

% bzr init-repo --no-trees qbzr
Shared repository (format: 2a)
Location:
shared repository: qbzr
% cd qbzr
% mkdir branches
% bzr branch lp:qbzr branches/trunk
Branched 1032 revision(s).
% bzr branch lp:qbzr/0.14 branches/0.14
Branches 969 revisions(s).
% bzr co --lightweight branches/trunk work
% cd work

After doing this, I can work commit/pull/push in the "work" directory as if I was in the "trunk" branch. Nothing exciting. Let's say I want to fix a bug in the "0.14" branch:

% bzr switch ../branches/0.14

Now I can work as if I was in the "0.14" branch. So I do some changes, commit them, do some more changes and realize that these should actually go to a new feature branch. So I'll not commit them and create the new branch (I use this very often, so I have branch --switch aliased to sbranch):

% bzr branch --switch ../branches/0.14 ../branches/new-feature

At this point the "work" directory points to the the "new-feature" branch and the uncommitted changes are still there. So I can commit them, do some more work, merge from other branches, etc. While working on something, I might want to run code from two branches at the same time for comparison. This is where Git doesn't help you, because you can have only one working tree at a time (unless you make a new clone of the repository). But with this layout in Bazaar, nothing says I can only have one checkout in the repository. I can actually have a checkout of one of the branches anywhere on the disk. So I do this:

% cd ..
% bzr co --lightweight ../branches/trunk tmp

And now I can run both versions from "work" and "tmp" side-by-side. After I'm done, I simply delete the "tmp" directory.

I'm writing this mostly because I'm surprised how little people know about it and I personally find it a very nice way to work in Bazaar.