How can we ensure Bazaar (bzr) remains active?

[Robert Collins]

On 22 September 2015 at 18:33, Eli Zaretskii <eliz at gnu.org> wrote:
>> Date: Tue, 22 Sep 2015 18:25:09 +1200
>> From: Robert Collins <robertc at robertcollins.net>
>> Cc: Jelmer Vernooij <jelmer at jelmer.uk>, Bazaar <bazaar at lists.canonical.com>,
>>       bronger at physik.rwth-aachen.de

>> Actually its in-principle possible to generate bzr format compressed
>> texts directly from git ones: it wouldn't be as compressed, but its a
>> fairly trivial transform.
>
> But then it won't be bzr-git, it will be something else, right?
>
> I was talking only about bzr-git as the engine behind this idea.

So bzr-git's job is 'pull from and push back to git'. That involves:

commit transformation (git revision id -> bzr revision ids, and the inverse)
tree transformation (generation of inventories for tree objects, and
the inverse)
content transformation (generation of file ids, and inventory chang
refs and creation of bzr compressed texts - and their inverses).

Changing the algorithm used by bzr-git to insert compressed texts in a
different fashion isn't sufficient to make it not bzr-git. A bigger,
and more important question would be whether the other operations
would be sufficiently cheap (perhaps assuming they are cached in a dot
directory alongside the .git dir) to be feasibly used as a UI to
access a git store).

I'm not sure it would be fast enough in all cases, but perhaps all
cases doesn't matter ? Like - bzr log -p isn't something I'd want to
do from a bzr store, if using bzr as a front-end to git. I'd want
instead to use bzr's history traversal algorithm, and then specify the
order to git itself - which is already very efficient at generating
its diffs.

-Rob

-- 
Robert Collins <rbtcollins at hp.com>
Distinguished Technologist
HP Converged Cloud