Bert Huijben wrote on Tue, Mar 29, 2016 at 15:05:21 +0200:
>
>
> > -----Original Message-----
> > From: kotkov_at_apache.org [mailto:kotkov_at_apache.org]
> > Sent: donderdag 24 maart 2016 12:59
> > To: commits_at_subversion.apache.org
> > Subject: svn commit: r1736432 -
> > /subversion/trunk/subversion/tests/cmdline/update_tests.py
> >
> > Author: kotkov
> > Date: Thu Mar 24 11:59:22 2016
> > New Revision: 1736432
> >
> > URL: http://svn.apache.org/viewvc?rev=1736432&view=rev
> > Log:
> > Fix a false negative in update_tests.py#76 with SVNPathAuthz short_circuit.
> >
> > This test used to assert that having a filename with backslash returns an
> > error during `svn update' over file:// or svn://, and doesn't do that over
> > http://. However, the last bit is specific to httpd instances configured
> > with SVNPathAuthz on, where a subrequest-based check gives 404 and
> > results
> > in the path being excluded by the server. With SVNPathAuthz short_circuit,
> > the behavior is the same as when the command is run over file:// or svn://
> > and results in an error:
> >
> > svn: E155000: 'completely\unusable\dir' is not valid as filename in [...]
> >
> > Since the original issue (SVN-3288) is about a crash, and given that we're
> > currently fine with either of the results, tweak the test to accept both of
> > them. There is an alternative of teaching the tests to distinguish between
> > different types of httpd authz and making the checks even more conditional,
> > but that seems a bit over the top.
>
> +1 on this change. Good fix.
>
> Perhaps we can extend it later on to verify in which mode we currently
> run the test suite (assuming the suite sets up the apache server)
We don't need detection to be complete, only to be sound.
davautocheck.sh sets up short-circuit mode if env['SVN_PATH_AUTHZ'] ==
'short_circuit', so we can have the test expect only the short_circuit
behaviour if that envvar is set to that value, and expect either
behaviour otherwise. (It seems that build/run_tests.py doesn't
currently support setting SVNPathAuthz.)
No opinion on whether we _should_ do this; just pointing out it's
possible.
Cheers,
Daniel
Received on 2016-03-29 21:44:49 CEST