On Wed, Feb 27, 2013 at 02:24:46PM +0800, Yiqiao Pu wrote:
> On 02/19/2013 10:02 PM, Eduardo Habkost wrote:
> >On Tue, Feb 19, 2013 at 06:05:01PM +0800, Yiqiao Pu wrote:
> >>On 02/15/2013 03:20 AM, Eduardo Habkost wrote:
> >>>On Thu, Feb 14, 2013 at 04:53:01PM -0200, Eduardo Habkost wrote:
> >>>>Hi,
> >>>>
> >>>>I am using the config file pasted at the end of this message. It is
> >>>>supposed to generate two test dictionaries for some test cases (the
> >>>>custom.* variants), one with KVM enabled ("kvm"), and one with KVM
> >>>>disabled ("nokvm"). But the result is:
> >>>>
> >>>>$ virttest/cartesian_config.py /tmp/cpuid.cfg
> >>>>dict 1: cpuid.default_vendor.kvm.cpu.unset.unknown_qemu
> >>>>dict 2: cpuid.custom.vendor.normal.kvm.cpu.unset.unknown_qemu
> >>>>$
> >>>>
> >>>>It looks like a bug to me. Or maybe I am missing some small detail?
> >>>Managed to reduce the test case even further:
> >>>
> >>>======================
> >>>variants:
> >>> - unknown_qemu:
> >>> - rhel64:
> >>>only unknown_qemu
> >>>variants:
> >>> - kvm:
> >>> - nokvm:
> >>>variants:
> >>> - testA:
> >>> nokvm:
> >>> no unknown_qemu
> >>> - testB:
> >>>=========================
> >>>
> >>>$ virttest/cartesian_config.py /tmp/cpuid.cfg
> >>>dict 1: testA.kvm.unknown_qemu
> >>>dict 2: testB.kvm.unknown_qemu
> >>>$
> >>>
> >>>testB.nokvm should have been skipped, but not testB.nokvm.
> >>>
> >>>Proof:
> >>>
> >>>$ virttest/cartesian_config.py /tmp/cpuid.cfg 'no testA'
> >>>dict 1: testB.kvm.unknown_qemu
> >>>dict 2: testB.nokvm.unknown_qemu
> >>>$
> >>>
> >>>
> >>Have a little test with your cfg file. And seems there is something
> >>wrong in the function might_pass() in get_dicts()
> >>It will return False when the content is empty. So maybe we need add
> >>a judgement for content in might_pass. Just add some line the end of
> >>that function like this, and I get the result we expect:
> >>
> >> if content:
> >> return False
> >> else:
> >> return True
> >What is the meaning of "content", exactly, and why should the function
> >return False if content is non-empty, and return True if it is empty?
> >
>
> Read the code again and seems the "content" is the current Condition
> object for the node. So if the "content" is empty, it means for this
> node we don't have any Condition object like filter and so on. So we
> should not failed this node in might_pass().
What about cases where content is empty, but we still want to optimize?
Aren't there such cases?
The main problem we have with this code is that just being correct is
not enough: we need it to be correct but still use a reasonable amount
of CPU and RAM to process our input. We can't make might_pass() return
True too often, or we risk making the optimization not effective
anymore, making the config parser a CPU and memory hog again.
(Right now, that code is still above my head and hard to reason about.
I would really love to see a description of the meaning of each
attribute/parameter/variable).
--
Eduardo
_______________________________________________
Virt-test-devel mailing list
[email protected]
https://www.redhat.com/mailman/listinfo/virt-test-devel
