[libvirt] [PATCH 19/20] qemuagenttest: Test arbitrary qemu commands and timeouting of commands

[Peter Krempa]

On 07/31/13 11:28, Daniel P. Berrange wrote:
> On Tue, Jul 30, 2013 at 01:01:56PM -0600, Eric Blake wrote:
>> On 07/30/2013 12:52 PM, Daniel P. Berrange wrote:
>>
>>>> Hmm, I wonder if it's worth adding some sort of escape hatch, maybe if
>>>> 'VIR_TEST_FAST' is defined to non-empty in the environment, then we skip
>>>> this test; developers that don't like the long wait can then export that
>>>> variable as 1, whereas the spec file can ensure it is 0.  That could be
>>>> a followup patch, though, and it might be worth getting more feedback
>>>> than just mine on whether the new long-running test needs tweaking to
>>>> allow developers to avoid waiting, while still avoiding bit-rotting of
>>>> the test at release time.
>>>
>>> IMHO we don't want any of the tests doing multi-second timeouts by
>>> default. IOW, rather than VIR_TEST_FAST, we should have a
>>> VIR_TEST_ALLOW_SLEEP=1, and ensure that libvirt.spec.in sets that
>>> when doing 'make check' and also make sure that autobuild.sh sets
>>> it. So all automated builds fully exercise the tests, but day to
>>> day usage isn't delayed
>>
>> For that matter, 'make check' for day-to-day usage should be able to
>> skip the gnulib subdirectory - the results in gnulib/tests will only
>> change if you upgrade the gnulib submodule, glibc, or some other core
>> component, which is not what we change on a day-to-day basis when
>> hacking gnulib, but is also something an autobuilder should be running
>> always.  I'll see if I can hack something up to speed up 'make check'
>> for normal users on the gnulib front, which we can then extend into
>> skipping Peter's new test.
>
> Good idea to skip gnulib tests.
>
>> GNU coreutils calls its variable RUN_EXPENSIVE_TESTS, defaulting to no,
>> but set to yes in autobuilders.  Sounds like the best type of naming
>> (maybe VIR_TEST_EXPENSIVE, to keep it in the VIR_ namespace).  Anyone
>> else want to chime in with a bikeshed color?
>
> That sounds like a fine name to me.

I like it too.

>
> Daniel
>

I pushed the rest of the series with addressing review comments except 
for this patch and I will now try to figure out the right way to skip 
expensive tests.

Thanks for your input.

Peter