On Tue, 27 Nov 2007, Paul Burba wrote:
> > -----Original Message-----
> > From: dglasser@gmail.com [mailto:dglasser@gmail.com] On
> > Behalf Of David Glasser
> > Sent: Monday, November 26, 2007 10:55 PM
> > To: Subversion Developers; Paul Burba
> > Subject: Inheritability of no_mergeinfo
> >
> > Should the dummy "no_mergeinfo" variable in
> > mergeinfo-sqlite-index.c have its inheritable flag TRUE or
> > FALSE? I think it would be best to be explicit.
>
> Seems it should be TRUE, since no_mergeinfo is used to represent
> svn:mergeinfo of "", which, having no ranges, is inherently inheritable.
> That said, I don't know that it matters right now what we set it to, I
> can't see that we use it anywhere. FWIW I ran merge_tests.py and
> merge_authz_tests.py over ra_svn using both TRUE and FALSE and it passed
> both ways.
>
> Really the question here is:
>
> "Is there a case where we need empty mergeinfo* '' to have the concept
> of inheritability?"
>
> I suppose there is, if we reverse merge a range to a target that would
> remove all mergeinfo from that target and the target has a child missing
> due to an authz restriction that wasn't missing when the first merge was
> performed. But this problem seems closely related to
> http://subversion.tigris.org/issues/show_bug.cgi?id=2881 and that is
> kinda dead in the water at the moment (at least as far as 1.5 is
> concerned). So for now I feel safe in setting no_mergeinfo's
> inheritability to TRUE and did that in r28068.
TRUE looks good.
FWIW, this likely got this way because the no_mergeinfo variable was added
before the inheritability flag, and was overlooked during the addition.
> * Until issue #3029 "Prevent mergeinfo with paths mapped to empty
> ranges" is resolved we could expand this question to include mergeinfo
> with paths mapped to empty ranges, e.g. "/trunk:".
- application/pgp-signature attachment: stored
Received on Tue Nov 27 18:32:20 2007