[master] 3ad1666 Remove the test of varnishtest -i with an empty argument.

[Dridi Boukelmoune]

On Thu, Dec 14, 2017 at 6:27 PM, Geoff Simmons <geoff at uplex.de> wrote:
>
> On 12/14/2017 05:38 PM, Dridi Boukelmoune wrote:
>
> > I have a workaround if we want to revive this check, let me know.
>
> Well, what does the workaround do?

Adds more arguments to the command line, only the last empty arg[s]
are accidentally stripped by libtool (and only on SunOS to my
knowledge).

    +shell -err -expect {-i: "" matches zero tags} \
    +       {varnishlog -i "" -g unreachable}

> The way I see it personally, we should automatically test as much of the
> code as we can, but if we do in fact test that path adequately with the
> vsl_glob_test, it doesn't seem urgent to restore the vtc test.

Testing from the outside ensures we catch usage regressions though.

>From the command line today:

    $ varnishlog -i "" -g unreachable
    -i: "" matches zero tags

> IIRC we were trying to get VTEST back to green, since some of the tests
> were failing with disturbing consistency.
>
> One of them, I think it was this one, resulted in the process table on
> the SunOS system being filled up with varnishtest zombies, which
> eventually led almost all of the tests failing.

That's a different problem.

> I remember spending a full day fighting with the shell quoting, until we
> finally gave up, and really don't want to have to do that again. But if
> there's a nice workaround that avoids the trouble, well sure.

Your mistake was to believe the quoting was wrong, I couldn't buy that
the shell working otherwise fine would fail this quoting only in
varnishtest. While I was trying to build hitch on different platforms,
after utterly failing to run the test suite on SunOS I figured I could
revisit this commit instead.

Dridi