Re: [PATCH 1/2] perf tools: Add reference timestamp to perf header

[Peter Zijlstra <peterz@infradead.org>]

On Mon, 2010-12-13 at 18:23 +0100, Frederic Weisbecker wrote:
> On Mon, Dec 13, 2010 at 06:13:39PM +0100, Peter Zijlstra wrote:
> > On Mon, 2010-12-13 at 18:11 +0100, Peter Zijlstra wrote:
> > > 
> > > Preferably, yes, but I don't see why we can't break the data file format
> > > if we've got good reasons to. 
> > 
> > I mean, we pretty much _have_ to break data file format when we want to
> > do splice() support.
> 
> Because we'll have one file per-cpu?

Right.

> But perf.data on UP will be sensibly the same as today so I suspect
> we won't need to be compatible.

Nope, since one file simply doesn't scale, data-streams will have to
live outside of the main file.

And when you have to make that happen, there is no reason to maintain
any kind of compatibility. The only thing you could do is provide some
script to convert old data files to the new format.

> But I guess I am missing something, in which case that probably doesn't
> change much the picture. It's not because one day we'll need to break
> the format that we can happily do so everyday. Backward compatibility
> is important and we should preserve it when it's possible.

But since you're all so focused on adding warts to the thing, maybe
possible has sailed. But see my email to acme.