Hyrum K. Wright wrote:
> 1.6.8 is a dud, and will never see the light of day. Thankfully, we
> found the problems with ra_serf and the deps during the testing, and
> appropriate fixes have been nominated.
>
> My plan is to reroll 1.6.9, but holding the roll until r900797 (a
> high-impact bug fix), and the 1.6.x-1.6.8-serfassertion branch are
> approved or vetoed. If possible, please review these two issues.
The regression when using serf is the only source code problem that
caused 1.6.8 to be rejected.
The purpose of r900797 is to make commits work properly through a DAV
write-through proxy server - so that they don't fail and need to be
re-tried if attempted at a time when the proxy is not up to date. I
understand that this is "high impact" because it happens very frequently
when using a busy repository through a mirror server.
The fix changes the way the client drives a commit over Neon, passing a
new flag and eliminating the PROPFIND. That seems to be a very sensible
and good thing to do, and was suggested by Greg in 2001, but it is a
protocol change and it has been in trunk only for approx 24 hours so
far. I would be happier if we can give it some wider exposure before we
consider it ready for back-port to a stable branch. In particular, if
anyone producing systems that connect in to Subversion's client-server
protocol, such as proxy servers and alternative servers, could test this
change, that would be reassuring. (My initial concern was for WANdisco's
products which are that kind of product, but I believe they will not be
affected by this particular change.)
Basically I am questioning whether r900797 should be seen as a
requirement for 1.6.9. Certainly it would be good to include this
additional fix if it's stable.
- Julian
Received on 2010-01-21 12:16:54 CET