Re: [tip:perf/core] tracing: Add DEFINE_EVENT(), DEFINE_SINGLE_EVENT() support to docbook

[Randy Dunlap <randy.dunlap@oracle.com>]

On Wed, 02 Dec 2009 09:55:35 -0500 Steven Rostedt wrote:

> On Wed, 2009-12-02 at 15:43 +0100, Ingo Molnar wrote:
> > * Steven Rostedt <rostedt@goodmis.org> wrote:
> > 
> > > On Wed, 2009-12-02 at 15:01 +0100, Ingo Molnar wrote:
> > > 
> > > > >  DECLARE_CLASS_AND_DEFINE_EVENT()
> > > > 
> > > > Hm, that's a bit too long. How about 'DEFINE_CLASS_EVENT()' as a 
> > > > compromise? It's similarly short-ish to TRACE_EVENT(), and it also 
> > > > conveys the fact that we create both a class and an event there.
> > > > 
> > > > The full series would thus be:
> > > > 
> > > > 	DECLARE_EVENT_CLASS
> > > > 	DEFINE_EVENT
> > > > 	DEFINE_CLASS_EVENT
> > > > 
> > > > hm?
> > > 
> > > I thought about that too, but it actually makes it more confusing. 
> > > Because, looking at this with a fresh POV, I would think that after I 
> > > declare a class, I would use DEFINE_CLASS_EVENT with that class.
> > 
> > yeah. Hence was my second-best choice 'DEFINE_STANDALONE_EVENT' or 
> > 'DEFINE_SINGLE_EVENT' - to stress the special nature it, and to actually 
> > nudge people towards creating classes of events instead of doing 
> > separate, standalone points. (which are a waste in the majority of 
> > cases)
> 
> But the current TRACE_EVENT is still defining a class. Thus, you could
> create a TRACE_EVENT (or whatever it is called) and then create
> DEFINE_EVENTs based on the TRACE_EVENT.
> 
> That's why I want a name that describes this.
> 
> DEFINE_EVENT_CLASS?
> 
> Perhaps that's the best.
> 
> DECLARE_EVENT_CLASS - only creates a class
> DEFINE_EVENT - defines an event based off of a class
> DEFINE_EVENT_CLASS - creates a class and defines an event by the same name
> 
> Perhaps this is best in keeping with linux kernel naming conventions?

Yes, that seems fairly typical.

I didn't care for the
DEFINE_x_y()
and
DEFINE_y_x()
suggestion.  That just opens the door for confusion IMO.

thanks,
---
~Randy