On 4 April 2014 20:45, Ben Reser <ben_at_reser.org> wrote:
>
> On 4/4/14, 5:02 AM, Ivan Zhakov wrote:
[...]
> > My opinion on Subversion 1.9.0 is the following: release Subversion
> > 1.9.0 ASAP without FSFS format change. We have many FSFS performance
> > improvements in trunk that doesn't require format change.
> >
> > I think that cost of maintaining disk format backward compatibility
> > and code destabilization doesn't worth the real benefits that users
> > get from fsfs7 performance improvements. On the other side: if
> > log-addressing and related stuff are so cool and rock-solid, users
> > always can switch to FSX and fully benefit from this new stuff.
>
> I guess all I can say is that I wish we'd had this discussion completely in
> November and resolved it. I'd like to see 1.9.0 in June (or earlier) but I
> don't know how realistic that is now that you want us to rip out significant
> amounts of work.
>
> Unfortunately though, none of this discussion helps with the issue at hand.
> What exactly do we need to do to make you happy with a 1.9.0-alpha release? Is
> there anything we can do? Is removing the format 7 code enough? Or if that
> was removed would you still be against it since you don't think it's useful?
>
> Bottom line, what do you want to do here?
Here are my thoughts.
I have the following concerns about the log addressing feature:
1. Data corruption is possible due to implementation bugs. We have had
a lot of bugs in revprop packing feature and I expect at least the
same amount of bugs in this feature. We are unable to identify these
bugs before the release due to obvious limitations of alpha and beta
testing. We haven't implemented the improved testing discussed in [1].
And overall quality of feature design and implementation seems far from
perfect.
2. Despite the possible bugs, maintenance burden will be increased
significatly. We are introducing additional FSFS format that
dramatically increase the code complexity. And we have a big code
duplication because this feature is actually copy-pasted from FSX
format.
3. I am not sure that the log adressing feature in the current
desing and implementaion is really valuable for users. On the other
side, we have the FSX format that is treated as experimental. The much
better way is to release log adressing feature in this new format and
see how it works in production (not during the alpha/beta testing).
These are my thoughts about the log adressing feature itself. Assuming
all of this I don't like the idea to release 1.9 alpha with this
feature. It will be easy to make an implicit decision for final
release "because we don't have any bug reports from alpha and beta
testers".
I propose to revert *all* the FSFS format changes (including the
svn_fs_move support) and then release the 1.9 alpha ASAP. I agree
that there are many other fixes and usefull improvements waiting for
release.
As far as I can tell, the svn_fs_move() is never called (quite surprising).
Moreover, I'm unable to find any unit tests for svn_fs_move(). So it's not
worth to change the repository format for non-used feature.
P.S. The whole situation reminds me the release of ra_serf in Subversion 1.7 :)
[1] http://mail-archives.apache.org/mod_mbox/subversion-dev/201311.mbox/%3CCABw-3YdV1YX0yU3cuWD8syPGpQxkLBUe=6h_bMkuBfA+vQf9XA@mail.gmail.com%3E
--
Ivan Zhakov
CTO | VisualSVN | http://www.visualsvn.com
Received on 2014-04-07 17:22:54 CEST