Brief article on benchmarks of Python repository with leading DVCSen

[Alexander Belchenko]

Barry Warsaw пишет:
> On Feb 11, 2009, at 10:45 AM, Paul Moore wrote:
> 
>>> Then they just have to branch all there branches below that directory.
> 
>> A purely personal comment - it's this second part that's the issue.
>> Bazaar is dictating my directory structure, and that's what I dislike.
> 
>>> It sounds to me like people are being completely unreasonable here.
> 
>> Actually, it sounds like the Bazaar proponents are repeatedly
>> misunderstanding the issue. That's OK, as long as you're not asking me
>> to change, if Bazaar suits you then there's no problem. Equally, if
>> the Bazaar developers are happy that the trade-offs make sense then
>> that's fine - not every tool can suit every user. But there *is* a
>> genuine conflict here.
> 
> Hi Paul.  I think you've mentioned this in the python-dev thread.  You
> have a point that a self-acclaimed adaptive dvcs should adapt to your
> preferred directory structure, but I'm one of those that doesn't really
> understand your problem.  Mostly that's because I really like the way
> Bazaar fits my own development style.
> 
>>> A fair benchmark would have been much more interesting IMHO. I expect
>>> Bazaar would
>>> have still be the slowest but it would at least be a reasonable and
>>> fair comparison.
> 
>> I don't want to get into benchmarking, but as a genuine use case, I
>> recently said I'd provide a small patch for Python. I used Mercurial,
>> and all I did was go to my "random junk" directory, do hg clone
>> http://,,,python... python-patch, cd python-patch, hack, then create &
>> upload the patch to the tracker.
> 
>> For Bazaar, I believe I'd need to add the steps
>> bzr init-repo python-repo
>> (do some fiddling to download and populate this repo with a pre-build
>> tarball)
>> (bzr pull the rest of the data from since the tarball was made)
>> bzr branch python-patch
>> ... etc much as before
> 
>> I don't know if the steps in parentheses are still needed, but they
>> certainly were, if you wanted to avoid extremely slow initial clone
>> times, at one point (and I haven't seen anything saying that's changed
>> - nope, I went and checked, it's still true that this is the
>> documented approach).
> 
>> That's certainly much more complex (not just "one command") - for a
>> one-off exercise where I intend to delete the whole lot after my patch
>> is accepted.
> 
> We know that the Bazaar repo on code.python.org is not in an efficient
> format, so I don't think its branch times can be completely trusted. 
> We'll fix that once we upgrade that machine or get closer to a real
> experiment.
> 
> Also, won't stacked branches make this much less painful?

I guess the answer: stacked branching is not default.