Re: [RFC 0/6] perf tools: Report event parsing errors

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

Em Mon, Apr 20, 2015 at 10:15:39PM +0200, Jiri Olsa escreveu:
> On Mon, Apr 20, 2015 at 05:09:36PM -0300, Arnaldo Carvalho de Melo wrote:
> > Em Sat, Apr 18, 2015 at 10:42:58PM +0200, Jiri Olsa escreveu:
> > So, do you want me to apply this series and you work on top of it or
> > should I wait for another that implements what Ingo suggested?
> 
> I'll resend v2 with the update
> 
> > 
> > Also, we have:
> > 
> > [root@zoo ~]# perf stat --all -e cycles usleep 1
> > 
> >  Performance counter stats for 'system wide':
> > 
> >          3,382,595      cycles                                                      
> > 
> >        0.001739102 seconds time elapsed
> > 
> > [root@zoo ~]# perf stat --all-cpus -e cycles usleep 1
> > 
> >  Performance counter stats for 'system wide':
> > 
> >          2,435,672      cycles                                                      
> > 
> >        0.001191689 seconds time elapsed
> > 
> > [root@zoo ~]# perf stat --a -e cycles usleep 1
> >   Error: Ambiguous option: a (could be --no-no-aggr or --append)
> > 
> >  usage: perf stat [<options>] [<command>]
> > 
> > [root@zoo ~]#
> > 
> > Which is nice, wouldn't it be good to have it for the that as well, i.e.
> > for your example "ev=" would be a valid, and shorter form of "even=" or
> > "event=".
> 
> hum, not sure I follow.. could you please explain in more details?

This would be considered valid:

    $ ./perf record -e 'cpu/even=0x1/' ls
    event syntax error: 'cpu/even=0x1/'
                             \___ unknown term
    valid terms: "event", "raw".

"even="  would be recognized as a shorter form of "event=", "evint="
would be invalid :-)

- Arnaldo