Michael Price <mprice@atl.lmco.com> writes:
> Philip Martin wrote:
> > In this particular case there are good arguments in favour of adding
> > the dump patch, for instance we cannot really change the DB schema
> > until we have released a tarball with a working dump. On the other
> > hand the dump/load code doesn't have much in the way of regression
> > tests.
>
> Well, right now the only major bug identified in 0.20 appears to be
> the dump/load bug. I have no problem making a release for
> that. However, if we are going to have a patch release for the
> dump/load code then perhaps more eyes should take a look at it to make
> sure other little issues haven't gone unnoticed.
In this particular case, the parser bug isn't like other nasty bugs,
whereby some complicated patch was made to a complicated chunk of
code, whose behavior could subtley change any number of other
behaviors.
This was a very simple change: the parser was outright missing a very
simple piece of logic: "before loading a new list of properties into a
filesystem node, erase the existing list first." That's the reason
property deletions weren't being honored by the parser.
AFAIK, I can't imagine any other codepaths that are likely to be
destabilized by this fix. Our dumping/loading code is a standalone
thing.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Fri Mar 21 22:19:37 2003