[Swan-dev] regarding: testing: adding missing : ==== end ==== to nicinit

[Andrew Cagney]

On Tue, 28 Jan 2020 at 14:31, Paul Wouters <paul at nohats.ca> wrote:
>
>
> > On Jan 28, 2020, at 18:45, Andrew Cagney <andrew.cagney at gmail.com> wrote:
> >
> >> On Tue, 28 Jan 2020 at 11:10, Antony Antony <antony at phenome.org> wrote:
> >>
> >> I am curious what your thoughts now?
> >> Is it a good idea to add " : ==== end ====" to nicinit.sh when final.sh is
> >> not necessary. Or just Antony's preference? The test author can decide?
> >
> > As an instrument I find it blunt.  It removes everything so useful
> > stuff run during the post-mortem, such as checking for a core file, is
> > lost.
>
> I see people using cut and paste and suddenly seeing multiple markers and missing markers. So I prefer to not use it if we can avoid them.
>
> > My preference would be to get away from final.sh (instead have
> > generic tear down code that, like swan-prep, runs silently unless
> > there's a real error).
>
> final.sh has really been overloaded when what was needed was something running on east after the run.sh. We have the numbered .sh method for that now (especially if support for that is added to namespaces)

I thought that worked.  The code's in fab/scripts.py.

> I do think things like checking for cores and audit logs can be done outside a .sh script.
>
> But shutdown is something I wouldn’t want there because often you want to ssh into the test. Although if kvmrunner also will support —shutdown (and thus not shutting down), than that could be useful (eg stock runs always having leak-detective reports)

The command:

   kvmrunner one-test

doesn't shutdown the domains.  The problem is that, to get and report
a valid test result, all the .sh scripts need to be run - including
final.sh and that unfortunately likes to kill libreswan.

My idea is to replace final.sh with a teardown script that, unless
there's a problem, leaves no trace in *.console.txt (but still spews
all over *.console.verbose.txt so we can tell it ran).  That way:

  kvmrunner one-test

can skip the teardown but still report a "valid" result.

If the teardown runs but doesn't make a sound ...