Re: Our failure on tracing tools unification (Was: Re: [RFC][PATCH00/15] tools: Unify perf and trace-cmd trace event format parsing v2)

[Frederic Weisbecker]

On Wed, Apr 25, 2012 at 10:05:25AM +0200, Ingo Molnar wrote:
> 
> * Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> 
> > On Mon, 2012-04-23 at 16:47 +0200, Frederic Weisbecker wrote:
> > > Ingo doesn't seem to want this library outside perf in order to avoid
> > > the fragmentation of the efforts on tracing tools.
> > 
> > I really don't see the point.. parsing 
> > /debug/tracing/events/*/format gunk really is separate, 
> > furthermore things like powertop et al already are separate 
> > projects and really need this lib.
> > 
> > I mean, who gives a bother where all that crap lives, if its 
> > in the kernel tree its all close enough to keep an eye on.. 
> > its not like kernel/events/ and kernel/trace/ are the same 
> > directory.
> > 
> > Ingo, please could you lighten up and let people get stuff 
> > done? Merging the two implementations, wherever the result 
> > lives, is a better situation that two dis-joint 
> > implementations. So either let Frederic work or do it yourself 
> > but don't hand-wave and road-block stuff.
> 
> As long as perf builds it not as a DSO - or has it in 
> tools/perf/libparseevent/ and co-installs the .so together with 
> perf (and it's packaged together) I'm fine with it.

It's a static library (.a) in tools/lib
This way we ensure perf is not going to dynamically link to some
random version. Besides it stays on the kernel tree so the sources
can evolve on the same branch.

The reason why it's on tools/lib is that people don't want to force
tools to depend on perf.

> One thing that would suck is perf being bound by various 
> versions of this library floating out there.

Sure, the fact it's static and is on the kernel tree prevents
from that.

> One reason why perf has become so popular is that it's 
> self-sufficient to a large degree and is very easy to 
> build/install. Lets not kill that aspect by the death of a 
> thousand cuts.

Agreed. The perf makefile just relays on tools/lib so the dependency
stays on the same tree and stays static so this is transparent to the
user that doesn't need to juggle with dependencies.

> 
> We had nothing but pain from external libraries so far, and 
> there's a reason why the Git project (3+ times larger than perf) 
> avoids library dependencies like the plague...
> 
> Anyway, Frederic, mind posting a version that does the above, so 
> I can have a look at how it all works in practice?

Ok I just rebased against latest perf/core updates and posted to
you. I also pushed it to some branch in my tree (see cover letter)

Thanks!

> 
> Thanks,
> 
> 	Ingo
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/