Re: [GIT PULL] perf update

[Frederic Weisbecker <fweisbec@gmail.com>]

On Thu, Jun 09, 2011 at 03:17:32PM +0200, Ingo Molnar wrote:
> > > 
> > >  - streamlined the renaming: we really want this to be ring_buffer.c 
> > >    (most of the complexity comes from this not being a simple buffer 
> > >    but a ring-buffer)
> > > 
> > >  - i streamlined the naming around it: struct ring_buffer 
> > >    internalized via internal.h (it does not clash with ftrace's 
> > >    ring-buffer)
> > > 
> > > It all looks and reads much nicer now, but please double check the 
> > > commit as well :-)
> > > 
> > > One other rename i'd like to do is:
> > > 
> > >    struct perf_output_handle		=> struct rb_handle
> > > 
> > >    perf_output_begin()			=> rb_open()
> > >    perf_output_copy()			=> rb_write()
> > >    perf_output_sample()			=> rb_write_sample()
> > >    perf_output_end()			=> rb_close()
> > > 
> > > Which really makes it a lot more apparent that it's a regular 
> > > input/output flow defined over the ring-buffer!
> > > 
> > > I can do this if this is fine with everyone. There will be no change 
> > > in functionality.
> > 
> > I feel more comfortable if we keep the perf_outpout_*() naming, having some
> > global rb_* would pollute the global namespace.
> 
> Hm, using the rb_ prefix is not good due to the (conceptual) clash 
> with rbtree.h primitives.
> 
> > perf_rb_* namespace would be fine as well.
> 
> How about:
> 
>     struct perf_output_handle		=> struct ring_buffer_handle
>  
>     perf_output_begin()			=> ring_buffer_open()
>     perf_output_copy()			=> ring_buffer_write()
>     perf_output_sample()		=> ring_buffer_write_sample()
>     perf_output_end()			=> ring_buffer_close()
> 
> ?
> 
> It doesn't clash with existing names.

For the same reasons I don't like the rename you did of the perf buffer into
struct ring_buffer, I think we shouldn't do this.

This is a generic naming that belongs to some very universal datatype. There
is no good reason to find that generic concept naming exported from a subsystem
that is not a generic implementation of that datatype.

And it's not because it doesn't yet clash with existing names that it's a good
choice.

Why not "struct perf_ring_buffer" ? And then perf_ring_buffer_begin()? or perf_rb_begin()
if it's too long.

> 
> Thanks,
> 
> 	Ingo