Re: [Qemu-devel] [PATCH v3 4/4] qemu-iotests: Test creating floppy drives

[John Snow]

On 10/19/2016 03:37 AM, Kevin Wolf wrote:
> Am 18.10.2016 um 21:53 hat Eric Blake geschrieben:
> > On 10/18/2016 02:45 PM, John Snow wrote:
> > > On 10/18/2016 06:22 AM, Kevin Wolf wrote:
> > > > This tests the different supported methods to create floppy drives and
> > > > how they interact.
> >
> > > > +function check_floppy_qtree()
> > > > +{
> > > > +    echo
> > > > +    echo Testing: "$@" | _filter_testdir
> > > > +
> > > > +    # QEMU_OPTIONS contains -nodefaults, we don't want that here
> > > > because the
> > > > +    # defaults are part of what should be checked here
> > > > +    echo "info qtree" |
> > > > +    QEMU_OPTIONS="" do_run_qemu "$@" | _filter_win32 |
> > > > +    grep -zo '[[:cntrl:]]\( *\)dev: isa-fdc.*\([[:cntrl:]]\1
> > > > .*\)*[[:cntrl:]] *dev:'
> > >
> > > This grep invocation doesn't appear to actually terminate with the '-z'
> > > option here. Not sure why, I haven't looked into the bash framework
> > > much, hopefully it's not too hard for you to reproduce and correct.
>
> No, obviously I can't reproduce, otherwise I wouldn't have written the
> test case like this. It passes just fine for me on RHEL 7.

Wasn't sure if it was something that popped up more recently or not. 
Obviously it worked at some point.

I'm on Fedora 24, using bash 4.3.42-7.fc24 and grep 2.25-1.fc24.

> Just to clarify, it's grep that doesn't terminate, or qemu? Also, what
> do you mean by the "bash framework"?

It seems like it's the grep invocation; I don't see any QEMU processes 
in `ps`, the only thing I can find is the grep invocation. (Why would 
grep hang if qemu has exited?)

By the 'bash framework' I meant the shell related infrastructure for 
iotests. I'm more familiar with the python parts.

> > Is 'grep -z' even portable to BSD, or is it just a GNU extension?  Would
> > it be better to run the output through tr to convert things to a saner
> > subset of characters before then grepping a text file?
>
> Is qemu-iotests even supposed to run on BSD? All our test cases specify
> "_supported_os Linux". (And I don't think this means Linux kernel with
> BSD userland :-))
>
> Anyway, the tr thing you mean would be translating all newlines into
> something else (which is hopefully otherwise unused), then grep, then
> translate back?
>
> What this line is supposed to do (if it wasn't obvious) is extracting
> the full information for a single device from 'info qtree'. I don't
> really mind how it's done, but multiline operation seems to be something
> that isn't trivial with most tools... I think I've done multiline sed
> before, so maybe that would be another option.
>
> Kevin