Re: [RFC PATCH 0/4] perf: Custom contexts

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

On Wed, 2011-03-16 at 14:53 -0300, Arnaldo Carvalho de Melo wrote:
> Em Thu, Mar 17, 2011 at 12:47:01AM +0900, Masami Hiramatsu escreveu:
> > (2011/03/16 10:03), Frederic Weisbecker wrote:
> > > Doh you're right. && would have two meaning.
> > > No we should probably keep  a && b  has a meaning of we are
> > > in the range a AND in the range b. Both at the same time, with
> > > a evaluated first and then b. We also need to ensure than
> > > a && b  doesn't mean the same than  b && a. You're right, perhaps
> > > we need another operator to expression inclusion, or we need to
> > > assume that specific meaning of &&.
> > > 
> > > For what I wanted to express in the example above, || seem be the
> > > right choice: -e inst*@(..lock:*acquire(B) || lock:*release(B)..)
> > > 
> > > So || would mean union and && would mean inclusion.
> > 
> > Hmm, would we really need that kind of complex rules?
> > It seems that we only need union case. If so, I'd suggest
> > you to use ',' to express that, instead of ||.
> > 
> > -e inst*@(..lock:*acquire(B),lock:*release(B)..)
> 
> Yeah, I somehow was avoiding the comma operator because it could be used
> to represent multiple events, but then its a different context, using it
> to represent a circular list of ranges in the @ (at, location) expression
> seems ok.
> 
> 1. '..lock:*acquire(B)' is armed, 'lock:*release(B)..' isn't
> 2. '..lock:*acquire(B)' trigers, which causes 'lock:*release(B)..' to be
>    armed
> 3. 'lock:*release(B)..' triggers, which causes '..lock:*acquire(B)' to
>    be armed, rinse, repeat

How about we start writing proper EBNF syntax rules for this stuff, its
getting seriously out of hand.