Paul Burba wrote: > Julian Foad wrote: >> I describe the idea of "logical changes" in >> <http://wiki.apache.org/subversion/MergeTrackingIdeas>. > > Most of the ideas there make sense, though I'm a bit foggy as to how > the mergeinfo format will change and how we'll migrate existing > repositories.
No, I only referred to that ideas page to explain the idea about categorizing changes as 'logical changes' versus 'merges'. I'm not proposing any mergeinfo format changes or anything that would need migration. > I know you are still early in the proof-of-concept phase, but have you > given much thought to performance? In particular I'm thinking of how > mergeinfo_graph_populate recursively scans the history of reflective > merges to build the mergeinfo graph. In doing so it ultimately calls > svn_client__get_repos_mergeinfo twice for each revision (operative or > not) described in the reflected mergeinfo. That could get slow in a > hurry. The 'twice' is obviously and simple avoidable. As for the rest, I've only thought a little as yet. In the worst case it is necessary, but in common cases I suspect we will be able to avoid much of it by delaying the graph population and scanning on demand, as we can prune the recursion at some points. - Julian