to group compress or not to group compress

[Talden]

On Thu, Jan 8, 2009 at 6:49 AM, Stefan Monnier <monnier at iro.umontreal.ca> wrote:
>> I don't think that is entirely true. It depends on the magnitude, but
>> people often use "this takes XXX MB" as a benchmark.
>
> Of course, they do, but current disk use of Bzr is good enough that
> improving it won't bring us many new converts.  Speed on the other hand
> is an issue that's brought up a lot more often by people who don't
> necessarily do comparisons with other VCS.

Au contraire. A brief test of replaying a subset of the commits we've
made in one of our smaller Subversion repositories suggests that a
Bazaar branch with all history would be 15-20% larger than the
equivalent Subversion space used... given the main Subversion
repository we'd really like to use Bazaar for is ~3GB that's a lot to
branch - now I expect as much as 10% of the Subversion repo is on
branches and wouldn't be pulled down but then again, we've used cheap
copy in a number of cases that would be reducing the total Subversion
size (this wasn't the case in the small subset tested).

For converts from centralised systems to decentralised systems size on
the client will be a factor and they're the main group Bazaar is
courting for converts.

Speed is important, but in many operations Bazaar and Subversion are similar.

The one place where Bazaar wins back some of that space is the shared
repository.  In Subversion I have 6-8 checkouts of the project
(mainline, release branches and dev branches) and these are very heavy
with working-copy meta-content (each is 1GB total)... There is
currently effort in Subversion to eliminate that but it will be a long
time until it is complete... Current space is about 6GB for 6
checkouts and greater than 6GB for bazaar with a shared repo - most
devs have one or two working copies and we'll likely get some
resistance from them, justified or not, when you tell them that 2GB
will become 4GB (they still point out that CVS was half as big as the
Subversion working copies).

> Yes, it's easy to do.  But if disk space is an issue, there's already
> a lot of options you can use to try and circumvent the problem: shared
> repositories, lightweight checkouts, stackable branches.  OTOH if "bzr
> annotate" is too slow for your taste, there's not much you can do.

I haven't looked into stacking yet - this could be the real life-saver.

Agreed about annotate - it's painfully slow in Subversion and you
don't even want to think about trying a revision graph for a file.

--
Talden