how to get bazaar log rid of minor "sandbox" commits?

[Rahul Nabar]

On Wed, Mar 11, 2009 at 7:04 PM, Martin Pool <mbp at sourcefrog.net> wrote:
> 2009/3/12 Rahul Nabar <rpnabar at gmail.com>:
>> branching does not work so well for me (I think) since bazaar
>> branches at the repo level. I am keeping 200 odd files in a single
>> repo located at ~/ But I often wish I could branch at file level but
>> the last I checked this was not a possibility. Or is there a way?
>
> Not really.  We feel it would get pretty complex if there were
> separate revision histories for different files within a directory -
> CVS does this and it can be complex enough there, and they don't
> handle consistency across files, changes to tree shape or merge
> tracking.

I did not like cvs so much for my purposes. I'd rather stick to bazaar
and live without this feature.

>
> It's conceivable using filtered views we could support composing a
> working tree from multiple branches.

That would be great.

> Basically what I would suggest is that you keep files that are
> logically related each in separate branches.

That would be sort of artificial and yet not much use. Imagine a
normal Linux box. Tons of config files, admin scripts, log plotting
snippets etc. I could , say, make my bin a seperate branch but it
would have about a 100 files.

Again if I need to "sandbox" the one file I am working on and decide
to branch it would needlessly involve the other 99 files in my bin
branch too. The limiting case, of course, is one-branch or one-repo
per file which is impractical.

I guess my problems are just because I am "hacking" Bazaar to suit my
targets. Inherently Bazaar (and other versioning systems as well; eg.
mercurial, subversion etc.) seems to be a project oriented system and
not really designed with the workflow of a "assorted sytem file config
versioning project."

Yet, its the best choice currently I feel (I have played with cfgmaker
etc. but they can't beat my current hacked up Bazaar system)

-- 
Rahul