On Tue, Oct 20, 2020 at 07:53:49AM +0200, Thomas Huth wrote: > On 16/10/2020 19.40, Andrew Jones wrote: > > On Fri, Oct 16, 2020 at 07:02:57PM +0200, Paolo Bonzini wrote: > >> On 15/10/20 18:35, Sean Christopherson wrote: > >>> The port80 test in particular is an absolute waste of time. > >>> > >> > >> True, OTOH it was meant as a benchmark. I think we can just delete it > >> or move it to vmexit. > >> > > > > If you want to keep the code, but only run it manually sometimes, > > then you can mark the test as nodefault. > > Please let's avoid that. Code that does not get run by default tends to > bitrot. I suggest to decrease the amount of loops by default, and if > somebody still wants to run this as a kind of benchmark, maybe the amount of > loops could be made configurable? (i.e. so that you could control it via an > argv[] parameter?) > I think both make sense. Making the number of loops variable is a good idea in order to keep the test running in CI in a reasonable amount of time (the timeout can also be adjusted by the CI runner, of course). Also, marking the test as nodefault makes sense if nobody really cares about the output unless they're specifically doing some benchmarking. Nothing stops travis or other CI from running nodefault tests, they just have to be explicitly requested. Thanks, drew
