On Mon, Apr 25, 2011 at 11:56 AM, Paul Burba <ptbu...@gmail.com> wrote: > On Sat, Apr 23, 2011 at 5:21 AM, Daniel Shahaf <d...@daniel.shahaf.name> > wrote: >> On Sat, 23 Apr 2011 07:15 +1000, "Daniel Becroft" <djcbecr...@gmail.com> >> wrote: >>> Hi, >>> >>> I've just updated to trunk@1096029, and have noticed the following change in >>> behavior. >>> >>> Previously, any reproduction script I write has the following pattern: >>> >>> svn checkout <url> workingcopy >>> > pushd workingcopy >>> > svn mkdir trunk >>> > svn commit >>> > svn switch <url>/trunk . >>> > >>> >>> This used to work fine. But now I'm getting the following error messages: >>> >>> subversion/svn/switch-cmd.c:184: (apr_err=195012) >>> > svn: E195012: Path '.' does not share common version control ancestry with >>> > the requested switch location. Use --ignore-ancestry to disable this >>> > check. >>> > subversion/libsvn_client/switch.c:201: (apr_err=195012) >>> > svn: E195012: 'file:///home/djcbecroft/dev/sandpit/repository/trunk' >>> > shares >>> > no common ancestry with '/home/djcbecroft/dev/sandpit/workingcopy' >>> > >>> >>> As the error message suggests, using --ignore-ancestry works fine. I don't >>> think this is required, though, as the paths share a common ancestry >>> already. >>> >> >> I've not read the whole thread, but tweaking your recipe I've just run into >> an assertion in a case I expected to work: >> >> [[[ >> % rm -rf wc1 r1 >> % $svnadmin create r1 >> % $svn co file://$PWD/r1 wc1 >> Checked out revision 0. >> % cd wc1 >> % $svn cp . trunk >> svn: E200007: Cannot copy path '/tmp/svn/wc1' into its own child >> '/tmp/svn/wc1/trunk' >> % $svn cp ^/ ^/trunk -mm > > Hi Daniel, > > Should we even allow a copy of r0? What exactly does it mean and how > is it useful? > > The assert you see here happens because mergeinfo has no concept of r0 > as a valid change. When we try this switch without using > --ignore-ancestry, switch.c:switch_internal() calls > svn_client__get_youngest_common_ancestor looking for common ancestor > of ^/trunk@1 and ^/@0. svn_client__get_youngest_common_ancestor() > uses svn_client__get_history_as_mergeinfo() to convert ^/trunk@1 and > /^@0 to svn_mergeinfo_t. > > Unfortunately svn_client__get_history_as_mergeinfo() does this > conversion by using svn_mergeinfo__mergeinfo_from_segments() to covert > svn_location_segment_t's into mergeinfo. > svn_location_segment_t.range_start can be the same as > svn_location_segment_t.range_end, but when > svn_mergeinfo__mergeinfo_from_segments converts these to > svn_merge_range_t, it enforces the rule that svn_merge_range_t.start < > svn_merge_range_t.end (see > http://svn.apache.org/viewvc?view=revision&revision=924201). > > So we end up with a bogus rangelist that later trips the assert when > svn_client__get_youngest_common_ancestor() sends it to the > svn_rangelist_intersect API. > > Unsure how to address this. We could tweak > svn_client__get_youngest_common_ancestor() to handle this edge case, > but I wonder if preventing copies of r0 in the first place is > something to consider too.
We allowed this in 1.6, so I made a few changes in r1096561 and r1906562 to continue supporting it. Paul
