1.0 roadmap

Martin Pool mbp at sourcefrog.net
Thu Nov 22 08:53:26 GMT 2007


On Nov 21, 2007 2:22 PM, Martin Albisetti <argentina at gmail.com> wrote:
> On Nov 20, 2007 3:29 PM, Nicholas Allen <allen at ableton.com> wrote:
> > It seems strange to me to have an almost ready high performance format
> > and not wait for it to mature before releasing 1.0. Otherwise users
> > will complain about performance of bazaar and if they think it is to
> > slow they may not give it another chance in the future. As packs seems
> > to solve many performance problems I would have thought 1.0 would
> > include it by default....
>
> I have to agree based on the feedback I get from users/potential users.
> I would very hard to explain why a high performance format is
> available and not on by default.
> It would plant the idea that "bzr just doesn't do well
> perfomance-wise" in everybody's head if 1.0 goes out without packs.
>
> Are there enough good reasons not to release with packs?
> And if there are, is the release of 1.0 *now* more important then
> delaying it a bit and having a much more complete 1.0 product?
>
> On the other hand, it seems the Windows issue needs some thought.
> Maybe leaving the Windows version in 0.9x until it's stable enough?
>
> And, on a slightly more positive note, bzr seems really solid here
> (linux environments, 15+ people using it, integration with custom
> software), and even without packs, performance is acceptable.
> I don't see any other reasons besides the ones mentioned above why 1.0
> shouldn't hit the streets.

I think you put it very well and I agree.

I think it would be kind of premature to do 1.0 with packs on the
verge of being ready, but not widely available.  What I would like to
do now is get them working well as the default format (mostly fixing
the bugs John listed), and release that as 1.0.

We've approved for landing a patch that makes the test suite pass
everything with pack repositories aside from hpss operations, and
hopefully can fix them tomorrow.  Reports from people testing packs
are positive.

We do need to do more on Windows, both technically and in the way we
handle it.  I feel like Alexander is a bit on his on, doing good work
but without much support from the rest for what is really a pretty
important feature.

Technically we need some core work to design and add in a way to
handle case-insensitive filesystems in the working tree.  It would be
nice if this code could be tested even on systems without a CIF.

But beyond that we have the problem that it's easy for tests to break
on Windows and accumulate so many failures that it's overwhelming for
Alexander or anyone to get them cleared.  To fix this I think we need
three things:

 1- Have someone put in the time to get all the tests on Windows clear
-- I'm willing to allocate Canonical people to this and John is
interested in doing it
 2- Get Windows tests running from pqm, so we can't check in things
that break it.  Robert suggests getting pqm to copy the merged tree to
a windows machine, ssh there and run the tests.  (There is a risk this
will slow pqm, as the suite is much slower on Windows.)
 3- Get some of the core developers dogfooding on Windows, at least
some of the time.  John or I might do it.  We have to accept paying a
price for doing it, not just in finding and fixing bugs, but also
fixing incidental things that make using and developing Bazaar on
Windows more reasonable -- maybe adding more gui operations, or
looking at better ssh integration on Windows.

Those are important and I'll sign up (myself and Canonical) to do
them, but they'll take some time - realistically a couple of months.
Because Bazaar on Linux is now pretty good, I think we should do 1.0
now, even though Windows support may not really be polished by itself.

-- 
Martin



More information about the bazaar mailing list