Re: [PATCH V3 3/5] perf,tools: Dump per-sample freq/CPU%/CORE_BUSY% in report -D

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

Em Wed, Jul 29, 2015 at 12:43:22PM -0300, Arnaldo Carvalho de Melo escreveu:
> Em Wed, Jul 29, 2015 at 02:53:29PM +0200, Jiri Olsa escreveu:
> > >  	switch (event->header.type) {
> > >  	case PERF_RECORD_SAMPLE:
> > > -		dump_sample(evsel, event, sample);
> > >  		if (evsel == NULL) {
> > >  			++evlist->stats.nr_unknown_id;
> > >  			return 0;
> > >  		}
> > > +		dump_sample(session, evlist, evsel, event, sample);
> > 
> > same here, you could pass only session all the way through
> 
> I'll take a look at how interesting it would be to have a evsel->evlist,
> that if NULL means the evsel is freestanding, but when it is linked to
> an evlist, then it will be there. This way we wouldn't have to pass
> (evlist, evsel) when the main purpose of a function is about an evsel
> but we need information that is logically associated to all evsels in a
> list, i.e. that is in evsel->evlist.

There are no cases where a function receives (evsel, evlist) with that
evlist containing that evsel :-\

Perhaps this will be the first, i.e. rename perf_session_env to
perf_env, then store it in evlist->env, then when processing something
where we have a evsel or evlist we can access that env from:

    evsel->evlist->env;

Will continue after lunch, trying to prototype what I just described.

What I have is in a tmp.perf/core branch in my tree..
  
- Arnaldo