On Sat, 27 Oct 2007, Daniel L. Rall wrote:
> On Fri, 26 Oct 2007, Senthil Kumaran S wrote:
>
> > Hi,
> >
> > I am attaching a patch along with this email in order to fix issue #2983.
> >
> > [[[
> > Fix issue #2983.
> >
> > When --parent-dir argument is passed to svnadmin, then prefix it
> > with the mergeinfo source paths.
> >
> > * subversion/libsvn_repos/load.c
> > (include): Include svn_mergeinfo.h
> > (prefix_mergeinfo_source_paths): New funtion to prefix mergeinfo
> > source path with parent_dir
> > (parse_property_block): If is_node is true and we have a parent dir
> > then prefix the mergeinfo source paths
> >
> > Patch by: Senthil Kumaran <senthil@collab.net>
> > ]]]
>
>
> This behavior makes a rather large assumption which I'm not entirely
> comfortable with (at least, not without some sort of option to disable it).
> We're basically assuming that all the merge source paths mentioned in
> svn:mergeinfo in the dump stream that we're loading is also in the dump
> stream. If it's not, and especially if it's present in other areas of the
> destination repository, this behavior is going to be screwing up any mergeinfo
> that is prefixed.
I hadn't realized this earlier, when Mike propsed this; I don't mean anything
else by it.
> Because the alternative of no mergeinfo adjustment seems worse, I've committed
> this patch to trunk with some tweaks in r27445.
It'd be really good to have a test case for this in svnadmin_tests.py.
There's already a test for 'load' in there, that could be used as a template
for a new test. In this case, a common helper function might be appropriate.
- application/pgp-signature attachment: stored
Received on Tue Oct 30 19:39:21 2007