Re: perf tool: About tests debug level

[Arnaldo Carvalho de Melo <acme@kernel.org>]

Em Tue, Jun 22, 2021 at 09:00:31AM -0700, Ian Rogers escreveu:
> On Tue, Jun 22, 2021 at 4:58 AM John Garry <john.garry@huawei.com> wrote:
> >
> > On 22/06/2021 06:04, Ian Rogers wrote:
> > >> ---- end ----
> > >> Parse and process metrics: FAILED!
> > >>
> > >> Note that the "FAILED" messages from the test code come from pr_debug().
> > >>
> > >> In a way, I feel that pr_debug()/err from the test is more important
> > >> than pr_debug() from the core code (when running a test).
> > >>
> > >> Any opinion on this or how to improve (if anyone agrees with me)? Or am
> > >> I missing something? Or is it not so important?
> > > Hi John,
> > >
> >
> > Hi Ian,
> >
> > > I think the issue is that in the parsing you don't know it's broken
> > > until something goes wrong. Putting everything on pr_err would cause
> > > spam in the not broken case.
> >
> > Right, I would not suggest using pr_err everywhere.
> >
> > > Improving the parsing error handling is a
> > > big task with lex and yacc to some extent getting in the way. Perhaps
> > > a middle way is to have a parameter to the parser that logs more, and
> > > recursively call this in the parser when parsing fails. I guess there
> > > is also a danger of a performance hit.
> >
> > So I am thinking that for running a test, -v means different levels logs
> > for test code and for core (non-test code). For example, -v prints
> > pr_warn() and higher for test logs, but nothing for core logs. And then
> > -vv for running a test gives pr_debug and above for test logs, and
> > pr_warn and above for core logs. Or something like that.
> >
> > Maybe that is not a good idea. But I'm just saying that it's hard to
> > debug currently at -v for tests.
> >
> > Thanks,
> > John
> 
> I think this sounds good. It'd be nice also to have verbose output in
> the shell tests following the same convention. There's currently no
> verbose logging in shell tests but I propose it here:
> https://lore.kernel.org/lkml/20210621215648.2991319-1-irogers@google.com/
> By their nature some of the shell tests launch perf, perhaps there can
> be some convention on passing the verbose flag through in those cases.

Hey, there is even a v2 for that one, lemme process it :-)

- Arnaldo