Bazaar repository size benchmarks

[Pieter de Bie]

On Sun, Jun 1, 2008 at 6:34 PM, Matthieu Moy <Matthieu.Moy at imag.fr> wrote:
> Actually, you imported only one branch for each project. It would be
> also interesting to see how the tools behave in the presence of many
> branches.
>
> I remember some discussion of Mercurial's list about some
> inefficiencies related to the presence of branches (the difficulty is:
> if you commit to branch A, then switch to branch B, and commit again,
> the best base to do a delta is not necessarily the last commited
> revision in the repository). I think Git would distance Mercurial even
> more, but the numbers would speak better than I ;-).

I have thought about this before discarding the branches. The reason I
did discard them is because I don't like the one branch / directory
approach, as branch management then becomes some sort of file
management. I couldn't find how to easily clean up a Bazaar repository
to regain lost space after deleting a branch, for example.

I think most performance attributes (size, speed) of branches should
still be in these repositories, as there are a lot of merges, and a
merge is basically just two branches joined by another commit. For Git
at least, there should be no size difference between having two
separate branches or having them joined in the same branch as a merge
(apart from the merge commit, of course). I think this is true for
Bazaar too, as it has a similar pack format (though I'm not familiar
with it). How Mercurial handles that, I don't know though. Perhaps it
would also depend on whether you use directory-branches or named
in-tree branches in Mercurial?

[ Matthieu, sorry for the double post, hit reply instead of reply-all ]