Re: possible bugfix found (Re: checkout_repository-1)

[Mark D. Baushke]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Paul Edwards <kerravon@nosppaam.w3.to> writes:

> > The patch as written will stop most of the tests due to "? paul.txt"
> > showing up in the various directories when a checkout is being done.
> 
> Ok, /tmp/paul.txt then.

Yeah, I did that and had a 

      [ -f /tmp/paul.txt ] && mv /tmp/paul.$what.txt

at the end of the case in the big "for what in $tests; do" loop.

> > All that said, I believe I have found the root-cause of the problem.
> 
> Great!

Well, it is hard to say if it worked or not. The Nightly testing using
the PowerMacintosh system that Derek has had some kind of a problem in
the Makefile that melted down... it seems unlrelated to my changes.

> > The problem appears to be that both ${CVS_DIRNAME}/CVS and
> > ${CVS_DIRNAME}/modules exist after the parseroot test is run.
> >
> > On MacOS X, /tmp/cvs-sanity/cvsroot and /tmp/cvs-sanity/CVSROOT are the
> > same directory... so, later, when the checkout_repository test is run,
> > it has problems. The solution would seem to be to run the parseroot test
> > inside of a subdirectory.
> >
> > I still need to finish running my regression tests before checkin, but
> > I think I have nailed this particular problem and I include the patch
> > after my .signature.
> >
> > If it works and the all of the nightly sanity tests pass, maybe we
> > can get Derek to release a new version of the feature branch.
> >
> > What I do not really understand is why this is not a problem in the
> > cvs-1-11-x-branch which seems to have the same parseroot test as the
> > trunk...
> 
> Also, why, when you run a range of tests, does this problem not
> occur.

There is only the ability to run a single test or to start at a
particular test and go to the end of all tests. Typically, I was backing
up only a few tests before the trouble and then running to completion
with no problems.

> Also, why did the error line effectively get corrupted, saying
> "? CVSROOT: failed to open directory"
> 
> instead of the prefix "cvs [checkout aborted]".
> 
> Why are two completely different error messages being
> merged in that instance?

A different set of code was being executed. The remote CVS saw that
there was already a CVS directory with all of the standard stuff one
might expect and called 'cvs update' instead of 'cvs checkout' ... To be
honest, there is a need for other tests in this area, so if anyone wants
to contribute to sanity.sh to test more features, that would be great.

        -- Mark
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (FreeBSD)

iD8DBQE/m3sm3x41pRYZE/gRAgq4AKCi2eZ0u/eX/PwRKuuOREYYfVw9tgCeMEeD
N895KU45QflhOaKwjpPdXwc=
=jnkz
-----END PGP SIGNATURE-----