> -----Original Message-----
> From: MARTIN PHILIP [mailto:codematt...@ntlworld.com] On Behalf Of
> Philip Martin
> Sent: dinsdag 22 januari 2013 18:56
> To: Bert Huijben
> Cc: dev@subversion.apache.org
> Subject: Re: upgrade_tests 29 XFAIL
>
> "Bert Huijben" <b...@qqmail.nl> writes:
>
> >> W: Couldn't find node 'E' in expected entries tree
> >> W: * Node name: E
> >> Path: svn-test-work/working_copies/upgrade_tests-29/A/B/E
> >> Contents: None
> >> Properties: {}
> >> Attributes: {'status': 'R ', 'wc_rev': '1'}
> >> Children: None (node is probably a file)
> >>
> >> Is the upgrade right or wrong to create the op-depth=3
> >> presence=incomplete row?
> >
> > I think the idea is that it adds this first and then when walking the
> > 'entries' in the directory itself the information is improved.
> >
> > Maybe we should have another set of operations to run when we don't
> find the
> > entries in the directory.
>
> I think with the above you are referring to the upgrade code? So you
> think that update should be changing the op-depth=3, presence=incomplete
> row into something else?
>
> > Another option would be to make the test verify that you can recover
from
> > this situation by running 'svn revert' on E. If that works it would be
good
> > enough for me.
> >
> > (The entries have a slightly different idea about what a replacement is
> > compared to status. Every node with a BASE node is replaced, even though
> the
> > node is not the op-root that replaces the node. There are entry_status,
> > entry_rev and entry_copied keywords to handle this in the test suite)
>
> As far as I can see tweak_for_entries_compare is removing the '!' node
> that the test puts into expected_status. I don't know how to make the
> test pass.
The entries code doesn't look to add disk state, so it never sees
obstructions or missing directory.
But currently entries-dump just exits on the assertion. So the output is
just ignored.
When reading the entries doesn't fail any more the test expectations can be
fixed by applying entry_status, entry_rev and entry_copied where necessary.
(If set these override the status types. See many of the tests I had to fix
last Saturday for some examples)
We have had this kind of overrides in the test suite, since before we got to
single-db.
Bert
