dbcheck failures on Fedora (was: Re: Remove (some) bashisms from the test scripts)

[Martin Schwenke]

On Mon, 29 Jan 2018 07:12:06 +1100, Andrew Bartlett via samba-technical
<samba-technical at lists.samba.org> wrote:

> On Sun, 2018-01-28 at 18:55 +0100, Timur I. Bakeyev wrote:

> > I was wondering myself what is the state of my patches? Do I need to do anything else to get them pushed into upstream?
> > 
> > Error you mentioned is a good example why we should eliminate bash-isms(or switch entirely to bash in the tests to avoid such problems).  
> 
> This one Andreas raised just before the release because this overall
> test now fails for him on Fedora, but not Ubuntu. 
> 
> We either need to do that (use bash explicitly) or need to have a way
> to ensure these don't come back. 
> 
> One approach would be to put ShellCheck (with an ignore pattern or
> expected set of errors) into make test, alongside the python pep8
> checks.  

I think that this is generally a good idea.  We did this to all the
non-test scripts in CTDB a while ago.

However, it was a lot of work, so I'd like to preempt any idea that we
mandate the use of ShellCheck.  It is a very blunt instrument that is
constantly changing.  The maintainers sometimes add checks that are of
limited value.  There was also a regression in version 0.4.5 (fixed in
0.4.7) but this is not yet in Debian (which is still at 0.4.6), so I
see see one test failure when I run on my laptop.

I think the approach should be to use ShellCheck on scripts that are
"known good" to ensure that we don't get any regressions.  I would love
for CTDB's test scripts to pass ShellCheck but that would require more
time than I'm willing to spend on it right now.

What I think ShellCheck desperately needs is a setting for the
level of compliance to be checked.  That way you could use it for
syntax checking and basic style, without having to check all of the
esoteric style violations.  Even better would be to have the checks
grouped in a way that you could exclude groups by name or by check
wildcard.

We would also need to decide what to do when ShellCheck is not
available.  Given that it is written in Haskell, it isn't the simplest
thing in the world to build... or patch.  For the CTDB shellcheck
test-suite we just print a message and pass the tests if ShellCheck
isn't installed.  This is probably what happens with our autobuilds on
sn-devel.

peace & happiness,
martin