On Mon, 03 Dec 2007, Paul Burba wrote:
> In r28213, I put some restrictions in place as to what values svn ps
> svn:mergeinfo will permit (previously we allowed just about anything).
> Most of these restrictions were already checked for in
> svn_mergeinfo_parse(), except for one: mergeinfo with paths that map to
> empty revisions. As
> http://subversion.tigris.org/issues/show_bug.cgi?id=3029* discusses
> these are redundant and best avoided. With r28213 The checks are done
> by leveraging svn_mergeinfo_parse() and then performing a second check
> for empty ranges.
>
> I'd like to move the empty range check into svn_mergeinfo_parse(), but I
> worry about users who have been working with trunk builds the last year
> and have mergeinfo in their repositories with paths mapped to empty
> ranges. Since svn_mergeinfo_parse() would error out when trying even
> list their svn:mergeinfo it would be pretty hard for them to fix. *So
> now for my real question*: How much deference should we give to those
> users who have been using trunk builds?
We make no compatibility guarantee on unreleased software. If there had
been an RC of 1.5, this might be a different story, but there hasn't; so
at this point, let's DTRT for the software.
> * Yes, yes, I know as we battle with issue #2897 and all the "Big" (tm)
> questions, that the small details like this don't get much attention,
> but a fantastic new A-class office building with the three L's ain't
> much if the toilets don't work.
We know for certain that the entire schema is going to change to properly
support reflected revisions (issue #2897). So, a database and/or FS-level
migration is already going to be necessary for early adopters.
--
Daniel Rall
- application/pgp-signature attachment: stored
Received on Mon Dec 3 21:38:01 2007