Today in IRC:
Aug 02 10:57:30 <Bert> pburba: When running tests with Single-DB I see
that on merging to missing files, the nodes are reported as missing..
but then some mergeinfo is recorded anyway. Is this the intended?
behavior (I think this was merge_tests.py 16)
Aug 02 10:57:52 <Bert> (that mergeinfo is recorded on the missing nodes)
Aug 02 11:00:22 <pburba> Bert: Yes, the missing file get's mergeinfo
representing what it inherited *before* the merge. So repeat merges
with the files present will notice that the files don't have the revs
from the first merge. We could do this without problem in WC-gen-1,
not sure if it is a problem with wcng. Is a test failing?
Aug 02 11:00:49 <pburba> Bert: If it is a problem, we can probably
work around it in other ways, setting non-inhertiable mergeinfo on the
parent instead.
Aug 02 11:04:45 <Bert> pburba: I just liked to know if it is a
problem. The notifications and on-disk status change for single db (as
only the in-wc file will be missing)
Aug 02 11:05:22 <Bert> So instead of a directory with an unknown
revision, it will be a missing directory with administrative data
available.
Aug 02 11:06:25 <pburba> Bert: Let's back up, is there a behavior
change with the move to a single db?
Aug 02 11:07:27 <pburba> Bert: And what is the change exactly?
Aug 02 11:07:28 <Bert> *nod* When you delete a directory... you don't
delete it's .svn
Aug 02 11:07:40 <Bert> So you still have its pristine files, properties, etc.
Aug 02 11:08:06 <Bert> pburba: 'rm dir', not 'svn rm dir'
Aug 02 11:11:10 <pburba> Bert: Sorry, I'm still not following what
your question is.
Aug 02 11:11:21 * pburba looks at merge test 16
Aug 02 11:13:34 <Bert> pburba: If you compile libsvn_wc with SINGLE_DB
and SVN_WC__SINGLE_DB defined, I see the recording of mergeinfo on
missing directories.. I was asking if this was expected behavior.
(Before single db, there was/is no place to store these properties; so
this is a behavior change)
Aug 02 11:19:53 <pburba> Bert: ok, I got you now. That change should
be an improvement, but let me recompile and check out that test. I'm
a bit foggy on the details of dealing with skipped-due-to-missing
targets and mergeinfo recording (it's been a while since I've been in
that particular space).
Bert,
What you saw was issue #2915, which is essentially about how we record
mergeinfo when subtrees are missing due to OS deletes. With a single
DB we now have a way of recording override mergeinfo on missing
directories (we've always handled missing files within a present
directory). I haven't quite settled my own mind on what the correct
behavior is here, but I dumped my present thoughts in
http://subversion.tigris.org/issues/show_bug.cgi?id=2915#desc4
Paul
Received on 2010-08-03 03:20:50 CEST