Hyrum K Wright <hyrum.wri...@wandisco.com> writes: > On Fri, Feb 10, 2012 at 4:24 PM, Bert Huijben <b...@qqmail.nl> wrote: >> >> >>> -----Original Message----- >>> From: hwri...@apache.org [mailto:hwri...@apache.org] >>> Sent: vrijdag 10 februari 2012 23:11 >>> To: comm...@subversion.apache.org >>> Subject: svn commit: r1242958 - >>> /subversion/trunk/subversion/libsvn_repos/log.c >>> >>> Author: hwright >>> Date: Fri Feb 10 22:10:34 2012 >>> New Revision: 1242958 >>> >>> URL: http://svn.apache.org/viewvc?rev=1242958&view=rev >>> Log: >>> Older (pre-1.6) revision files don't contain the information to fully >>> populate >>> the changed_path hash for 'log -v' and 'log --xml'. If we encounter such a >>> condition, fetch the kind manually. >> >> Why do we add this now? >> >> We started sending the information 'when available' with 1.6, and I >> think we didn't retrieve the data for performance reasons. Did >> something change? > > User request (through private channels). They were parsing the output > of 'log --xml' which shows a blank 'kind' attribute against these old > revisions. Fetching the kind as part of the log is much easier, both > for the user and for performance, than individually running 'svn info' > on each node. > > Arguably, they could cache this information somewhere client-side, > since the historic information isn't going to change. But that's what > the VCS is for in the first place. > >> This might be quite a performance regression for 'svnadmin upgrade'-d >> servers that were from <=1.5 before the last dump-load and it won't >> make the information magically available on clients, because there is >> no requirement to upgrade the server. > > Possibly, but I think we already say that running 'log -v' may take a > while longer than "normal" log. I also think the number of such > instances likely to be encountered in the wild are few, and the > benefits to those that do want this information outweigh the costs.
So this code didn't work for deleted nodes, it's necessary to track back to a revision in which the node exists. I've fixed it with r1245800 but that probably makes it now even more expensive. -- uberSVN: Apache Subversion Made Easy http://www.uberSVN.com
