bzr-colo and shelve

[Brian de Alwis]

Hi Neil.

On 15-Apr-2010, at 8:24 PM, Neil Martinsen-Burrell wrote:
> Can you say more about the workflow that combines bzr-colo with CVS?
> I'm not familiar enough with CVS (thank goodness) to know how it helps.
> I'm just curious about how bzr-colo in particular helps in that situation.


I use bzr + colo to maintain a parallel but private set of branches to manage my own work.  The branches are short-lived.

CVS, like Subversion, places a "CVS" directory in every directory throughout the tree to record information about the version-controlled files in that directory.  I exclude those CVS/ directories from the bzr repository but leave them in place, thus enabling the use of 'cvs diff' or Eclipse's CVS Team support to generate patches of the current branch against HEAD.  The Eclipse CVS Team patches are augmented with additional Eclipse-specific information that simplifies applying patches from within Eclipse (the patches are relative to each project/module's root, rather than the checkout's root), and is the preferred way to submit patches.

One alternative was to add the CVS/ directories to the bzr tree so that every branch had the CVS information; since my changes are relatively short-lived, this was more work than really necessary, and I wanted to avoid pain from merging CVS-specific information.  Or I could maintain a pristine trunk, merge my changes from a work branch, create the patch, and then revert.

(It may well be that bzr-eclipse can generate the Team-style patches; I didn't pursue this as (i) I'm so used to bzr on the command line that I haven't tried bzr-eclipse yet, and (ii) it seemed very unlikely since the Eclipse Team stuff is deeply tied to CVS.)

Thanks for your work on bzr-colo, BTW.  It reflects very well on bzr's architecture that such a plugin is even possible!

Brian.