Re: [PATCH] perf tools: fix write_event()

[Peter Zijlstra <peterz@infradead.org>]

On Sun, 2010-01-31 at 09:32 +0100, Ingo Molnar wrote:
> * Xiao Guangrong <xiaoguangrong@cn.fujitsu.com> wrote:
> 
> > We parse the event while it's read from mmap buffer in
> > write_event(), but sometimes the event will straddles the
> > mmap boundary, we should handle this case
> > 
> > And if we record events(such as perf kmem/sched) for long
> > times, Ctrl + C can't interrupt it just for this reason
> > 
> > Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
> > ---
> >  tools/perf/builtin-record.c |   62 +++++++++++++++++++++++++++++++++---------
> 
> Note, i've reverted the original instead, via:
> 
>   a8e6f73: Revert "perf record: Intercept all events"
> 
> As Peter said it, it is fundamentally wrong for perf record to touch the data 
> stream in the common case. Arnaldo, cannot we use some other method for perf 
> archive's needs? Worst-case we can do an optional parsing of the data stream 
> although we should first try hard to find a generic method ...

Arnaldo posted patches for this last friday somewhere, it might turn out
you already found them (reading mail by date order here).