With this change, our code no longer reads "all entries", nor does it attempt to cache that hash of entries.
The only way to do so, is with svn_wc_entries_read(), svn_wc_walk_entries3(), or svn_wc_entry(). All three are deprecated and not used by trunk. Not seeing much of a performance gain, though. (I suspect SQLite caching was helping us before) Cheers, -g ps. this breaks special_tests 12 and stat_tests 5. I'll fix after sleep... On Wed, Apr 21, 2010 at 06:04, <gst...@apache.org> wrote: > Author: gstein > Date: Wed Apr 21 10:04:42 2010 > New Revision: 936240 > > URL: http://svn.apache.org/viewvc?rev=936240&view=rev > Log: > Revamp and simplify the central entry-fetching function, > svn_wc__get_entry. It no longer worries about the entries cache, and > always goes to the database to fetch the parent/entry pair of entries. > > * subversion/libsvn_wc/entries.c: > (svn_wc__get_entry): strip out the access baton and entries cacheing > considerations from this function. use read_entry_pair to grab the > target entry. only use the RESULT_POOL and SCRATCH_POOL rather than an > access baton's pool. > > Modified: > subversion/trunk/subversion/libsvn_wc/entries.c >...
