Re: [PATCH] perf tools: Add 'G' and 'H' modifiers to event parsing

[Jiri Olsa <jolsa@redhat.com>]

On Tue, Apr 17, 2012 at 12:37:13PM +0300, Gleb Natapov wrote:
> On Mon, Apr 16, 2012 at 10:10:30PM +0200, Jiri Olsa wrote:
> > On Sun, Apr 15, 2012 at 06:23:00PM +0300, Gleb Natapov wrote:
> > > On Sun, Apr 15, 2012 at 05:12:22PM +0200, Peter Zijlstra wrote:
> > > > On Sun, 2012-04-15 at 17:59 +0300, Gleb Natapov wrote:
> > > > > -modifier_event [ukhp]{1,5}
> > > > > +modifier_event [ukhpGH]{1,5} 
> > > > 
> > > > You'd also need to increase the 5 to 7 I think, unless there's a clear
> > > > exclusion with some of the other flags.
> > > > 
> > > > Hmm, as it stands I think the 5 is already too low: "ukhppp" is 6.
> > > > 
> > > > We should really write this as ['u'] ['k'] ['h'] ['G'] ['H'] ['p'{1,3}],
> > > > not the combined thing.
> > 
> > you mean to match them strictly in parser, right?
> > because what we have now can match also stuff like 'uuu' 'kkuu' ...
> > 
> > So far it eludes me how to do that purely in flex and
> > doing this via rules in bison seems like overkill.
> > 
> > I'd rather keep it the way we have and do the extra check
> > in parse_events_modifier function.. unless someone comes up
> > with flex regular expression that would do that (which is
> > possible given my regex knowledge..) 
> > 
> > also we need 8 characters for it, since this seems to be
> > valid modifier: ukhGHppp
> > 
> So want me to resend with 8 to fix missing GH for now? I can't think of
> proper regex either.

That would be great.. any chance you could add a test to the builtin-test.c? ;)
Or just modify current one (one of the test__parse_events tests) to use those
flags, should be straightforward..

thanks,
jirka