Re: [PATCH 1/3] rbtree: Introduce rb_for_each_entry

[Frederic Weisbecker <fweisbec@gmail.com>]

On Wed, Jan 13, 2010 at 03:18:16PM +0100, Peter Zijlstra wrote:
> On Wed, 2010-01-13 at 12:13 -0200, Arnaldo Carvalho de Melo wrote:
> > Em Wed, Jan 13, 2010 at 02:58:53PM +0100, Peter Zijlstra escreveu:
> > > On Wed, 2010-01-13 at 11:52 -0200, Arnaldo Carvalho de Melo wrote:
> > > > Em Wed, Jan 13, 2010 at 02:34:45PM +0100, Frederic Weisbecker escreveu:
> > > > > At least it matches many patterns in perf :)
> > > > 
> > > > What an insane piece of... software! 8-) Whatever, if it makes Peter
> > > > happy we can keep it on tools/perf/util/include/linux/rbtree.h, just
> > > > after the:
> > > > 
> > > > #include "../../../../include/linux/rbtree.h"
> > > > 
> > > > ;-)
> > > > 
> > > > Just like we have list_del_range() in tools/perf/util/include/linux/list.h.
> > > 
> > > That's not the point, doing a for_each on the rb-trees as we have today
> > > is a O(n log(n)) operation, not something you should want to do often.
> > > 
> > > Adding a helper promotes the idea that its a sane thing to do, its not.
> > > 
> > > If you really need it, open coding it isn't hard, but the lack of helper
> > > does make you think and hopefully realize you're doing something funny.
> > 
> > We need it in several places in the perf tools, to present sorted results, to
> > dump the maps for debugging purposes, etc.
> > 
> > But I'll go and look at each one of them to see if there is any where it
> > is used in some stupid way.
> 
> Sure, but again, that's missing the point, adding that helper isn't a
> good thing. We have no 64bit division operators in Linux either, for the
> very same reason.
> 


I don't understand why this is unsane. Sure it is O(n log(n)), but
that's what is suitable for our use in perf. Shouldn't we simplify
it through a helper just because it's not something sane in most
other cases?

But yeah, such helper inside the kernel wouldn't be sane, as it
wouldn't fit any sane need I guess.