Stacked branching question

[David Ingamells]

Colin D Bennett wrote:
> On Mon, 21 Jul 2008 18:20:10 +0200
> David Ingamells <david.ingamells at mapscape.eu> wrote:
>
>   
>>  My point is exactly that there should be enough information in a
>> shallow branch to be able to do a diff and thus derive a delta.
>>
>> That is what I meant by a _little_ bit of depth. The little bit of
>> depth being one revision deep. (In other word not zero revisions
>> deep). It is exactly from the use-case you mentioned that I am coming
>> from. If you think about what I said should be possible (e.g diff on
>> local changes)
>>
>> it is implicit that there should be something local to diff on (and
>> thus provide a basis for the deltas you say are essential)
>>     
>
> But in some cases you *don't* want to duplicate past revisions.  Say
> you are stacking on another local branch, as you might on a code
> hosting server:  then you are trying to save space by not duplicating
> data on the server.
>
> And it would be nice to be able to use the entire revision history to
> get an optimum delta, rather than only having the past few revisions to
> base it on.
>
> Regards,
> Colin
>
>
>   
I do not exclude the option to have a zero-depth, what I am saying is 
that what I am waiting for is the special type of stacked branching 
called "shallow" branching, where the branch has just a little depth. 
Ideally for me the branch command would have an option --depth n where 
the last n revisions are saved in the branch.

Are you saying that bzr can use revision data from, say, file Y in the 
delta of file X? I doubt that for the work I see coming through from my 
team that a delta against just the last revision will be far from 
optimal. I think user would rather have a quick branch and commit and 
lose a small amount of disk space in the main repos, which I would wager 
is only a couple of percent.