Re: GNOME DVCS Survey Results

["Elijah Newren" <newren gmail com>]

Hi,

[Disclaimer: I wasn't involved in the construction or running of the
survey, other than the analysis you saw plus some late feedback on the
survey questions (I think my feedback was merely to suggest the
"other" answer for contributor types.)]


2009/1/5 Andrew Cowie <andrew operationaldynamics com>:
> On Sat, 2009-01-03 at 22:46 -0500, Behdad Esfahbod wrote:
>> GNOME contributors with an SVN account who had an SSH key installed on
>> their account were invited to fill in the survey.
>
> [It is NOT my intention to get all negative here; I understand - and
> accept - that projects make decisions and not everyone is happy with
> them. Luckily this decision ultimately is one I can ignore.
> Nevertheless, I have been asked by a number of people to write to this
> thread with why I am so dissatisfied. I do appreciate the effort people
> made, even if I feel that the way the whole survey exercise was
> conducted it was impossible for Git to lose]

Thanks for taking the time to do so.  I'm sorry you feel that way
about the survey exercise; I was encouraged that people were moving
forward and that they had created what I felt was an unbiased survey
as possible (Behdad asked for feedback from the foundation board,
release team, and others before sending it out, and I think my main
comment at the time was that I was happy to see the lack of bias in
the survey; my only other comment was the "other" contributor thing,
IIRC.)

> Some comments:
>
> ++
>
> It's a shame that hackers who contribute to GNOME projects which don't
> use svn.gnome.org were excluded.
>
>        (I was told their opinions didn't matter. {shrug} that's fine,
>        so long as nobody tries to represent this survey as "what GNOME
>        hackers think")

How would you draw the line?  Who should be included and who
shouldn't?  And how do we contact them all?  I think doing a survey of
any group other than those with svn commit access would be practically
unmanageable...and far more likely to be suspected of
non-representative-ness [New word!].

Also, if users aren't using GNOME svn then why would they care if we
switch or not?  Shouldn't the survey poll those whom it would affect?

> ++
>
> It was also a shame that I (one who does happen to have a GNOME svn
> account) was not able to complete the survey either because it crashes
> Epiphany when you i) vote for bzr and ii) withhold your vote from git,
> hg, and svn.
>
>        (When I asked if it might be possible to fix the survey so that
>        GNOME's web browser didn't crash, I was told "known bug" and
>        "too bad, you have to express a preference for Git and Mercurial
>        even if you don't want to". Strange take on democracy. I am
>        rather accustomed to the idea that declining to express a
>        preference for something is an acceptable form of voting.
>        Whatever)
>
> I explicitly did not want to chose Git or Mercurial, because I knew
> exactly what was going to happen. I've heard it several times already in
> #gnome-hackers and elsewhere:
>
>        "so it seems the people who prefer Bazaar like Git as their
>        second choice, so surely it's ok to go with that. Great!
>        Decision made"
>
> No. The rest of the survey was irrelevant. It was quite evident that the
> object of the exercise was to allow people to say "lots of people said
> Git was either their first or second choice" which sounds very
> impressive, and was exactly the one thing I did NOT want to support.
>
> So it crashed my browser. Nice.

That sucks.  Big time.

However, you'll be happy to know that most of my analyzing work was
originally performed on an alternative output file that included all
partial answers.  (It also contained a bit more data, such as svn
usernames -- and thus I can verify that your response was included in
this file and just did so.)  Now the reason this is relevant is that
when I got the final data in an alternate format from Behdad, I had
already generated all my plots and written my analysis.  So I had to
regenerate all the plots and compare old and new versions.  It turns
out the two sets were basically indistinguishable to my eye.  I spot
checked a couple of my claims in my analysis (e.g. that translators
preferred git over svn since it was so close in both data sets), but
actually didn't check them all.  Thus, you could say that my analysis
is more valid (or at least more verified) for the set of users that
also includes partial answers like yours.

On a related note, it looks like the total number of people who ranked
the various systems (taken from the data file including partial
answers) are:
  any: 581
  bzr: 585
  git: 583
  hg: 581
  svn: 582
So, yes, it looks like there were more people who left git unranked
than bzr -- the difference being two people.


However, I'm a bit confused by this statement:

   No. The rest of the survey was irrelevant. It was quite evident that the
   object of the exercise was to allow people to say "lots of people said
   Git was either their first or second choice" which sounds very
   impressive, and was exactly the one thing I did NOT want to support.

So vote for git fourth or fifth instead of second?  You didn't have to
vote for it first or second.

Also, if the point of the survey was as you claim then I totally
missed it -- you'll notice that I was more interested in first and
*last* place rankings in my analysis (and looked at
second-to-last-place rankings when I noticed that "any" was filling up
last place).  The reason for that was that I assumed that git was
divisive (since there are a number of people who vocally claim it's so
horrible -- which has included me in the past), and hence that it
would get a lot of last place votes.  But it didn't.  And I'm still
shocked.

> ++
>
> We chose the Bazaar decentralized version control system for our GNOME
> project even before the people behin GNOME's centralized code hosting
> made the courageous and monumental decision to switch from CVS to
> Subversion. Since GNOME didn't offer any way for us to host our
> 'mainline' branch on any official sounding resource, {shrug} we didn't.
>
> And so we don't.
>
> And that's actually the only issue that matters so far as I can tell. No
> one can force us to stop using Bazaar. People who work at places like
> Immendio who are using Git to hack on GTK+ cannot be forced to stop git
> either.
>
> And I wouldn't want them to. They're happy with their tool. We're happy
> with ours.
>
> When CVS was the only interchange (actually, that's not true, since the
> real interchange for most projects is attachments to Bugzilla of all
> things), then indeed "GNOME switching to Subversion" was a big deal.
>
> But in the era of distributed version control, the next step really
> matters little. Whatever GNOME _infrastructure_ offers "next" in terms
> of hosting is really quite irrelevant, since quite anyone can host their
> own projects and publish their own branches with nothing more than a
> vanilla web server.
>
> If the choice had happened to be Bazaar, then we probably would have
> moved our principle copy of our 'mainline' branch there. That would have
> been "nice" but otherwise is inconsequential since hosting the primary
> 'mainline' somewhere else costs us nothing, and I long since offered
> other people accounts to publish their own branches there too. But since
> it's going to be Git, well, it offers nothing for us.
>
> If the choice had been the other way around, then Git people would
> simply continue to host their branches somewhere else as they already
> area. Again, no change.
>
> This is ultimately why the whole debate is a bit pointless.

The lower the stakes the greater the venom.

Sorry, I couldn't resist.  :-)