Re: Auto Upgrade Behavior

On Mon, Aug 27, 2012 at 1:58 AM, Bert Huijben <bert_at_qqmail.nl> wrote:
>
>
>> -----Original Message-----
>> From: Stefan Sperling [mailto:stsp_at_elego.de]
>> Sent: zondag 26 augustus 2012 21:39
>> To: Greg Stein
>> Cc: Branko ÄŒibej; dev_at_subversion.apache.org
>> Subject: Re: Auto Upgrade Behavior
>>
>> On Sun, Aug 26, 2012 at 02:29:45PM -0400, Greg Stein wrote:
>> > On Aug 25, 2012 8:08 PM, "Branko ÄŒibej" <brane_at_wandisco.com> wrote:
>> > >
>> > > On 26.08.2012 00:31, Greg Stein wrote:
>> > > > In the past, we used auto-upgrade because it "just worked". Most
>> > > > users don't need or want to worry about working copy formats. They
>> > > > just want
>> > svn
>> > > > to work.
>> > > >
>> > > > I don't think we should be making things more difficult for the
>> > majority in
>> > > > order to help a few users who use multiple clients. That is backwards.
>> > :-(
>> > >
>> > > Well, evidence appears to suggest that users who use multiple
>> > > clients are in fact the majority. Hearsay evidence, but that's the
>> > > only kind I see hereabouts.
>> >
>> > I'd call it a vocal minority. We've got millions of users. I can't see
>> > the majority using multiple clients. Nobody runs into issues using a
>> > single client, so there is no need to speak up.
>>
>> I keep getting these complaints often. Mostly from users I personally talk to
>> during workshops, consulting, etc. Dunno how much of our user population
>> they represent. However it would be nice for me and them if auto-upgrade
>> was disabled by default. Because the problem would be solved for them, and
>> I could spend the time during my workshops talking about more interesting
>> stuff than why we auto-upgrade and how to avoid the pitfalls. My desire to
>> make this changed is based on real user feedback. I wouldn't want to make it
>> if these people didn't request it.
>
> For AnkhSVN I got complaints about auto upgrades for 1.5 -> 1.6, but I got far more complaints about the not auto upgrades to 1.7.

Weren't those complaints mostly about various upgrade problems (bugs
in upgrade code or subtle 1.6-corruptions which only appeared during
upgrade-to-1.7), rather than about the non-automaticness? The upgrade
to 1.7 was far more complex than any other wc upgrade before, which of
course yielded many more problems. From my point of view that made it
even more of a relief that it wasn't automatic.

> I don't think the users really know about the problems of the other choice. They just see a problem with the way they use it now and say that it should be the other way. They haven't tried it the other way and checked if this would have caused bigger other problems.

Agreed, it's quite hard to get an accurate picture here.

> Most AnkhSVN users have multiple clients, but we educated them for 1.5, 1.6 that they should upgrade clients like TortoiseSVN at the same time and for my users that is not a problem.

Hm, but that's because you can perfectly align AnkhSVN's release
schedule with the main release of svn (and TSVN does as well). If all
of a user's tools are catching up to the svn release very quickly,
then there is indeed not much of a problem (except if users don't like
the new release for some reason (perhaps upgrade-related problems),
and they'd rather wait some more for a more stable patch release). But
release-alignment of all the tools isn't always the case, and we
shouldn't put our users automatically into a dead end in that case.

> As Greg noted: The discussion on these topics on our list is really about the vocal minorities. If you just look at reports on our mailing lists the upgrade for 1.6 to 1.7 seldom works well, while in practice at least hundreds of thousands of users performed this upgrade -probably each on multiple working copies- without any problems.
>

Well, when I look at a fixed population, like my user base (which I
admit is very small (50 developers of various levels)), most of them
were very happy (relieved) with the non-automatic upgrade of 1.7.
That's not vocal minorities, that's me looking at the same group of
users and comparing between the 1.5->1.6 transition and the 1.6->1.7
transition. Granted, it's a small sample, but it's the only
information I have :-) ...

-- 
Johan