Lightweight branches?

[Martin Pool]

On 15 Nov 2006, Aaron Bentley <aaron.bentley at utoronto.ca> wrote:
> Ross Burton wrote:
> 
> > I had a look at the history horizon spec, and its exactly what I want.
> > How hard would it be to implement?

It will be somewhat challenging as it touches some deep levels but if
you're keen I can at least tell you how to start preparing for it.

> I think it would be time-consuming to implement.  It requires a new
> Repository format, changing the code that copies data from one
> repository to another, and probably changing some assumptions in bzr.

The "collation" storage format we discussed in Sydney goes towards this
in a couple of ways, and I've started some code for it.  

There are two assumptions our storage layer makes that (it seems to me)
need to be relaxed to do history horizons.  They are that there's a
single knit file per file_id, and that the storage layer's indexing is
overloaded to also describe the ancestry graph.  Instead we should be
able to group together historic data more flexibly so we can have old
history in a separate location or not present at all.

Another example of this is that knits are required to store accurate
annotations.

-- 
Martin